CN114387450B - Picture feature extraction method and device, storage medium and computer equipment - Google Patents
Picture feature extraction method and device, storage medium and computer equipment Download PDFInfo
- Publication number
- CN114387450B CN114387450B CN202210026806.2A CN202210026806A CN114387450B CN 114387450 B CN114387450 B CN 114387450B CN 202210026806 A CN202210026806 A CN 202210026806A CN 114387450 B CN114387450 B CN 114387450B
- Authority
- CN
- China
- Prior art keywords
- picture
- feature vector
- sub
- picture feature
- pictures
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a picture feature extraction method, a device, a storage medium and computer equipment, relates to the technical field of information, and mainly aims to improve the precision of picture feature extraction. The method comprises the following steps: acquiring a picture to be processed in an actual service scene; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures based on the first picture feature vectors and the position information of the plurality of sub-pictures in the to-be-processed picture respectively; and determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector. The method is suitable for extracting the picture characteristics.
Description
Technical Field
The present invention relates to the field of information technologies, and in particular, to a method and apparatus for extracting a picture feature, a storage medium, and a computer device.
Background
The picture feature extraction is a basic task in natural language processing, and provides a basis for subsequent natural language processing through the picture feature extraction, so that the efficient and accurate picture feature extraction has important significance for the natural language processing.
At present, a feature extraction model is generally constructed by using clear and complete pictures, and picture feature extraction is performed based on the constructed feature extraction model. However, in an actual service scenario, the picture is often incomplete or unclear, for example, a cut picture or a mosaic picture, which results in a feature extraction model constructed by using a clear and complete picture, and it is difficult to support feature extraction of the unclear or incomplete picture, and the extraction accuracy of the picture features cannot be guaranteed.
Disclosure of Invention
The invention provides a picture feature extraction method, a picture feature extraction device, a storage medium and computer equipment, which mainly aim at improving the precision of picture feature extraction.
According to a first aspect of the present invention, there is provided a picture feature extraction method, comprising:
acquiring a picture to be processed in an actual service scene;
Dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value;
Inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture;
determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed;
And determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
According to a second aspect of the present invention, there is provided a picture feature extraction apparatus comprising:
the acquisition unit is used for acquiring the picture to be processed in the actual service scene;
the dividing unit is used for dividing the picture to be processed into a plurality of sub-pictures and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value;
the extraction unit is used for inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction, so as to obtain a first picture feature vector corresponding to the target sub-picture;
the first determining unit is used for determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the to-be-processed picture respectively;
And the second determining unit is used for determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
According to a third aspect of the present invention, there is provided a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
acquiring a picture to be processed in an actual service scene;
Dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value;
Inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture;
determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed;
And determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
According to a fourth aspect of the present invention there is provided a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of:
acquiring a picture to be processed in an actual service scene;
Dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value;
Inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture;
determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed;
And determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
According to the picture feature extraction method, the picture feature extraction device, the storage medium and the computer equipment, compared with the mode that a feature extraction model is built by using clear and complete pictures at present and picture feature extraction is carried out based on the built feature extraction model, the picture feature extraction method and the picture feature extraction device are used for obtaining pictures to be processed in an actual service scene; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed; determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, thereby determining a clear target sub-picture in the picture to be processed by dividing the picture to be processed, extracting the first picture feature vector corresponding to the target sub-picture, determining second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures based on the first picture feature vector and the position information of the plurality of sub-pictures in the picture to be processed respectively, and finally determining the third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, so that the situation that feature extraction is performed on an unclear picture by using a feature extraction model constructed by clear and complete pictures can be avoided, and the precision of picture feature extraction is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:
Fig. 1 shows a flowchart of a picture feature extraction method provided by an embodiment of the present invention;
Fig. 2 shows a flowchart of another picture feature extraction method according to an embodiment of the present invention;
Fig. 3 is a schematic structural diagram of a picture feature extraction device according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of another image feature extraction device according to an embodiment of the present invention;
Fig. 5 shows a schematic physical structure of a computer device according to an embodiment of the present invention.
Detailed Description
The application will be described in detail hereinafter with reference to the drawings in conjunction with embodiments. It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other.
At present, a feature extraction model is built by using clear and complete pictures, and the picture features are extracted based on the built feature extraction model, because the pictures in the actual service scene are often incomplete or unclear, the feature extraction model built by using the clear and complete pictures is difficult to support feature extraction of the unclear or incomplete pictures, and the extraction precision of the picture features cannot be ensured.
In order to solve the above problems, an embodiment of the present invention provides a method for extracting image features, as shown in fig. 1, where the method includes:
101. And obtaining a picture to be processed in the actual service scene.
The pictures to be processed in the actual service scene comprise pictures which are subjected to mosaic, cut pictures and the like.
For the embodiment of the invention, in order to overcome the defect of lower precision of picture feature extraction in the prior art, the embodiment of the invention divides the picture to be processed into a plurality of sub-pictures, determines a clear target sub-picture from the plurality of sub-pictures, simultaneously, extracts a first picture feature vector corresponding to the target sub-picture by using a picture feature extraction model, determines a second picture feature vector corresponding to the rest sub-pictures in the plurality of sub-pictures based on the first picture feature vector and position information corresponding to the plurality of sub-pictures, and finally determines a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, thereby avoiding the situation that feature extraction of an unclear or incomplete picture is difficult to support by using a feature extraction model constructed by clear and complete picture. The embodiment of the invention is mainly applied to a scene for extracting the picture characteristics, and the execution main body of the embodiment of the invention is a device or equipment capable of extracting the picture characteristics, and can be arranged at the side of a client or a server.
Specifically, a large number of pictures to be processed exist in an actual service scene, the pictures to be processed can be obtained in a picture database by downloading the cut pictures or the pictures with mosaics on a network, after the pictures to be processed are obtained, the pictures to be processed are divided into a plurality of sub-pictures, clear target sub-pictures are determined from the plurality of sub-pictures, meanwhile, the target sub-pictures are input into a preset picture feature extraction model for feature extraction, a first picture feature vector corresponding to the target sub-pictures is obtained, then second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures are determined based on the position information of the first picture feature vector and the plurality of sub-pictures in the pictures to be processed respectively, and finally third picture feature vectors corresponding to the pictures to be processed are determined based on the first picture feature vector and the second picture feature vector, so that the picture feature extraction precision is improved.
102. Dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value.
For the embodiment of the present invention, if the sizes of the plurality of sub-pictures are the same, the to-be-processed picture is divided into a plurality of sub-pictures, and the number of the plurality of sub-pictures is as follows:
N is the number of sub-pictures, a is the length corresponding to the sub-pictures, b is the width corresponding to the sub-pictures, H is the length corresponding to the picture to be processed, W is the width corresponding to the picture to be processed, after the picture to be processed is divided into a plurality of sub-pictures, clear target sub-pictures are determined in the plurality of sub-pictures, the number of the clear target sub-pictures can be 1 or more, the application is not particularly limited, after the clear target sub-pictures are determined, the target sub-pictures are input into a preset picture feature extraction model for feature extraction, a first picture feature vector corresponding to the target sub-pictures is obtained, and based on the position information of the first picture feature vector and the plurality of sub-pictures in the picture to be processed, second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures are determined, finally, based on the first picture feature vector and the second picture feature vector, the feature extraction precision of the third picture corresponding to the picture to be processed is ensured.
103. And inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture.
The preset picture feature extraction model may specifically be a preset encoder.
For the embodiment of the invention, after the clear target sub-picture in the picture to be processed is determined, the target sub-picture is input into a preset encoder for feature extraction, a first picture feature vector corresponding to the target sub-picture is obtained, and based on the position information of the first picture feature vector and the plurality of sub-pictures in the picture to be processed respectively, a second picture feature vector corresponding to the rest sub-pictures in the plurality of sub-pictures is determined, and finally, a third picture feature vector corresponding to the picture to be processed is determined based on the first picture feature vector and the second picture feature vector, so that the situation that feature extraction is carried out on an unclear picture by using a feature extraction model constructed by clear and complete pictures can be avoided, and the precision of picture feature extraction is improved.
104. And determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed.
The position information may specifically be a sequence of positions of the plurality of sub-pictures in the to-be-processed picture, and the plurality of sub-pictures are encoded into 1,2, 3, and the like sequentially according to the sequence of the plurality of sub-pictures from left to right and from top to bottom in the to-be-processed picture, so that the position information of the plurality of sub-pictures in the to-be-processed picture can be determined, where the position information includes position information corresponding to the target sub-picture.
For the embodiment of the invention, after determining the first picture feature vector corresponding to the target sub-picture, in order to determine the third picture feature vector corresponding to the to-be-processed picture, first, determining the second picture feature vector corresponding to the rest sub-pictures except the target sub-picture in the plurality of sub-pictures, based on the second picture feature vector, the method comprises determining the position information of the plurality of sub-pictures in the to-be-processed picture respectively, based on the first picture feature vector corresponding to the target sub-picture and the position information of the plurality of sub-pictures in the to-be-processed picture respectively, and determining a second picture feature vector corresponding to the rest of the sub-pictures in the plurality of sub-pictures, and finally determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, so that the condition that the feature extraction model constructed by the clear and complete picture is utilized to extract the features of the unclear picture can be avoided, and the precision of the feature extraction of the picture is improved.
105. And determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
For the embodiment of the invention, after the first picture feature vector corresponding to the target sub-picture in the picture to be processed and the second picture feature vector corresponding to the residual sub-picture are obtained, the first picture feature vector and the second picture feature vector are finally added to obtain the third picture feature vector corresponding to the picture to be processed, so that the situation that the clear target sub-picture in the picture to be processed is determined by dividing the picture to be processed, the first picture feature vector corresponding to the target sub-picture is extracted, and meanwhile, the second picture feature vector corresponding to the residual sub-picture after the target sub-picture is removed in the multiple sub-pictures is determined based on the position information of the first picture feature vector and the multiple sub-pictures in the picture to be processed, and finally, the third picture feature vector corresponding to the picture to be processed is determined based on the first picture feature vector and the second picture feature vector, so that the condition that the feature extraction model constructed by clear and complete picture is used for carrying out feature extraction on the picture is avoided, and the feature extraction precision of the picture feature extraction is improved.
According to the picture feature extraction method provided by the invention, compared with the mode that a feature extraction model is constructed by using clear and complete pictures at present and picture feature extraction is carried out based on the constructed feature extraction model, the picture to be processed in an actual service scene is obtained; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed; determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, thereby determining a clear target sub-picture in the picture to be processed by dividing the picture to be processed, extracting the first picture feature vector corresponding to the target sub-picture, determining second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures based on the first picture feature vector and the position information of the plurality of sub-pictures in the picture to be processed respectively, and finally determining the third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, so that the situation that feature extraction is performed on an unclear picture by using a feature extraction model constructed by clear and complete pictures can be avoided, and the precision of picture feature extraction is improved.
Further, in order to better illustrate the above process of extracting the picture features, as a refinement and extension of the above embodiment, another method for extracting the picture features is provided in the embodiment of the present invention, as shown in fig. 2, where the method includes:
201. and obtaining a picture to be processed in the actual service scene.
Specifically, a picture database contains a large number of pictures which are demosaiced or damaged in an actual service scene, in order to obtain picture feature vectors corresponding to a picture to be processed, the picture to be processed can be obtained in the picture database, the picture to be processed is segmented to obtain a plurality of sub-pictures corresponding to the picture to be processed, clear target sub-pictures are determined in the plurality of sub-pictures, then the target sub-pictures are input into a preset picture feature extraction model for feature extraction, a first picture feature vector corresponding to the target sub-pictures is obtained, second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures are determined based on position information corresponding to the first picture feature vector and the plurality of sub-pictures respectively, and finally a third picture feature vector corresponding to the picture to be processed is determined based on the first picture feature vector and the second picture feature vector.
202. Dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value.
Specifically, the number of divisions is determined first, then the picture to be processed is divided based on the number of divisions, for example, if the number of divisions is determined to be 25, the picture to be processed can be divided into 5 parts transversely, then divided into 5 parts longitudinally, the picture to be processed is divided according to the number of divisions, then a plurality of sub-pictures can be obtained, a fresh target sub-picture is determined in the plurality of sub-pictures, then the target sub-picture is input into a preset picture feature extraction model for feature extraction, a first picture feature vector corresponding to the target sub-picture is obtained, meanwhile, a second picture feature vector corresponding to the rest sub-pictures in the plurality of sub-pictures is determined based on the first picture feature vector and the second picture feature vector, and finally a third picture feature vector corresponding to the picture to be processed is determined based on the first picture feature vector and the second picture feature vector.
203. And determining a pixel matrix corresponding to the target sub-picture.
Specifically, the target sub-picture is converted into a pixel matrix, if the picture to be processed is a color picture, the target sub-picture is converted into an RGB pixel matrix of a×b×c, wherein a is the length of the matrix, b is the width of the matrix, c is the channel number of the matrix, the pixel value at each position in the pixel matrix is a numerical value between 0 and 255, in order to reduce the calculated amount of the model, each pixel value needs to be divided by 255, that is, the pixel matrix is subjected to normalization processing, so that the size of each pixel value is between 0 and 1, and accordingly, a normalized pixel matrix corresponding to the target sub-picture can be obtained, a fourth picture feature vector corresponding to the target sub-picture is determined based on the normalized pixel matrix, then the fourth picture feature vector is input into a preset picture feature extraction model, a first picture feature vector corresponding to the target sub-picture is obtained, and simultaneously, a second picture feature vector corresponding to the second picture is determined based on the first picture feature vector and the residual feature vector corresponding to the second picture feature vector, and a final feature vector corresponding to the second picture is determined based on the residual feature vector.
204. And transversely splicing all rows of pixels in the pixel matrix to obtain a fourth picture feature vector corresponding to the target sub-picture.
Specifically, after determining a pixel matrix corresponding to a target sub-picture, transversely splicing each row of pixels in the pixel matrix by using reshape functions to obtain a feature vector of a first preset dimension corresponding to the target sub-picture, for example, if the dimension of the normalized pixel matrix is p×p×c, where C represents the number of channels of the pixel matrix corresponding to the target sub-picture, converting the matrix of which the dimension is p×p×c into a vector of which the dimension is P2C by using reshape functions, and using the method to obtain the feature vector of the first preset dimension corresponding to the target sub-picture, further, in order to meet the precision of feature extraction performed by a preset feature extraction model, converting the feature vector of the first preset dimension into a feature vector of a second preset dimension, namely a feature vector of a fourth picture, where a specific conversion formula is as follows:
Y＝WX
In the method, in order to extract a first picture feature vector corresponding to a target sub-picture by using a preset picture feature extraction model, the second preset dimension can be set to 1024, then finally the feature vector of the first preset dimension corresponding to the target sub-picture is converted to a fourth picture feature vector of 1024 dimension, and the fourth picture feature vector corresponding to the target sub-picture is input into the preset picture feature extraction model for feature extraction, so as to obtain the first picture feature vector corresponding to the target sub-picture, and based on the first picture feature vector and the position information of the plurality of sub-pictures in the picture to be processed, the second picture feature vector corresponding to the residual sub-picture in the plurality of sub-pictures is determined, and finally the third picture feature vector corresponding to the picture to be processed is determined based on the first picture feature vector and the second picture feature vector.
205. And inputting the fourth picture feature vector into the preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture.
The preset image feature extraction model may specifically be a preset encoder, where the preset encoder includes an attention layer and a feedforward neural network layer, the output of the attention layer is used as input of the feedforward neural network layer, and the attention layer includes a plurality of attention subspaces, that is, the attention layer in the present application adopts a multi-head attention mechanism, and further, in order to improve the accuracy of image feature extraction, 24 preset encoders connected in series are used to extract features of a target sub-image, each encoder is connected end to end, and the output of the previous encoder is used as input of the next encoder to extract feature vectors in the target sub-image.
For the embodiment of the present invention, before inputting the target sub-picture to a preset encoder, in order to improve the accuracy of extracting the feature vector by the preset encoder, the preset encoder needs to be trained in advance, and based on this, the method includes: acquiring a sample picture in an actual service scene and an actual sample picture feature vector corresponding to the sample picture; dividing the sample picture into a plurality of sample sub-pictures, and determining a clear target sample sub-picture from the plurality of sample sub-pictures; inputting the target sample sub-picture into an initial picture feature extraction model to perform feature extraction to obtain a first sample picture feature vector corresponding to the target sample sub-picture; determining a second sample picture feature vector corresponding to the sample picture based on the first sample picture feature vector and the position information of the plurality of sample sub-pictures in the sample picture respectively; constructing a loss function corresponding to the initial picture feature extraction model based on the actual sample picture feature vector corresponding to the sample picture and the second sample picture feature vector; training the initial picture feature extraction model based on the loss function, and constructing the preset picture feature extraction model.
Specifically, firstly, obtaining a sample picture in an actual service scene, carrying out segmentation processing on the sample picture to obtain a plurality of sample sub-pictures, determining a clear target sample sub-picture in the plurality of sample sub-pictures, inputting the clear target sample sub-picture into an initial picture feature extraction model to carry out feature extraction to obtain a first sample picture feature vector corresponding to the target sample sub-picture, then determining a third sample picture feature vector corresponding to the rest sample sub-pictures in the plurality of sample sub-pictures based on the first sample picture feature vector and the position information of the plurality of sample sub-pictures in the sample picture, finally determining a second sample picture feature vector corresponding to the sample picture based on the first sample picture feature vector and the third sample picture feature vector, and constructing a loss function corresponding to the initial picture feature extraction model based on the actual sample picture feature vector corresponding to the sample picture and the second sample picture feature vector, wherein the method specifically comprises the steps of: calculating each vector difference at the same position in the actual sample picture feature vector and the second sample picture feature vector; and constructing a loss function corresponding to the initial picture feature extraction model by calculating the square sum of the vector differences.
Specifically, after determining an actual sample picture feature vector and a second sample picture feature vector corresponding to the sample picture, calculating each vector difference at the same position in the actual sample picture feature vector and the second sample picture feature vector, and then squaring and summing the vector differences to obtain, that is, calculating a root mean square error corresponding to the initial picture feature extraction model, and constructing a loss function corresponding to the initial picture feature extraction model by calculating the root mean square error, where a formula for specifically calculating the root mean square error is as follows:
Wherein Z represents root mean square error, u1, u2...ur represents each vector in the second sample picture feature vector, v1, v2...vr represents each vector in the actual sample picture feature vector, r represents the number of vectors corresponding to the actual sample picture feature vector, after constructing a loss function corresponding to the initial picture feature extraction model according to the above formula, training the initial picture feature extraction model based on the loss function until the minimum loss function value appears, and constructing the preset picture feature extraction model based on model parameters corresponding to the minimum loss function value, namely the preset encoder in the embodiment of the invention.
Further, after obtaining the preset encoder with better performance through training, the first picture feature vector corresponding to the target sub-picture needs to be extracted based on the encoder, based on which, step 205 specifically includes: inputting the fourth picture feature vector into different attention subspaces in the attention layer for feature extraction to obtain a fifth picture feature vector of the target sub-picture in the different attention subspaces; multiplying and summing a fifth picture feature vector of the target sub-picture in the different attention subspaces with weights corresponding to the different attention subspaces to obtain an attention layer output vector corresponding to the target sub-picture; adding the attention layer output vector and the fourth picture feature vector to obtain a sixth picture feature vector corresponding to the target sub-picture; and inputting the sixth picture feature vector into the feedforward neural network layer to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture.
Specifically, the fourth picture feature vector is input to different attention subspaces in the attention layer of the first encoder to perform feature extraction to obtain a fifth picture feature vector of the target sub-picture in the different attention subspaces, wherein the process of specifically obtaining the fifth picture feature vector of the target sub-picture in the different attention subspaces is to determine a query vector, a key vector and a value vector of the target sub-picture in the different attention subspaces according to the fourth picture feature vector, multiply the query vector of the first target sub-picture in the different attention subspaces with the key vector of the target sub-picture in the different attention subspaces to obtain the attention score of the target sub-picture in the different attention subspaces for the first target sub-picture, and meanwhile multiply and sum the attention score of the target sub-picture in the different attention subspaces with the key vector to obtain the fifth picture feature vector of the first target sub-picture in the different attention subspaces, so that the fifth picture can be obtained as to obtain the fifth picture feature vector of the target sub-picture in the different attention subspaces.
Further, the fifth picture feature vector of the target sub-picture in the different attention subspaces is multiplied by the weights of the fifth picture feature vector and the weights of the sixth picture feature vector in the different attention subspaces and summed to obtain the output vector of the attention layer in the first encoder, meanwhile, the output vector of the first encoder and the fourth picture feature vector residual are added to obtain the sixth picture feature vector corresponding to the target sub-picture, the sixth picture feature vector is then input into the feedforward neural network layer to obtain the output vector of the feedforward neural network layer, the output vector of the feedforward neural network layer and the input vector of the feedforward neural network layer, namely the sixth picture feature vector residual are added to obtain the output vector of the first encoder, and the output vector of the first encoder is input into the second encoder to conduct feature extraction in a serial mode, the output vector of the second encoder is used as the input vector of the next encoder, the output vector of the first encoder is finally used as the input vector of the next encoder, the output vector of the second encoder is finally used as the input vector of the second encoder, the final picture is determined based on the position of the first picture, the final picture is determined based on the first picture, the final picture is the final picture, and the final picture is determined based on the final feature vector of the first picture and the final picture is determined.
206. And determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed.
For the embodiment of the present invention, after determining the first picture feature vector corresponding to the target sub-picture, the second picture feature vector corresponding to the remaining sub-pictures in the plurality of sub-pictures needs to be determined based on the first picture feature vector and the position information of the plurality of sub-pictures in the to-be-processed picture, based on which step 206 specifically includes: and inputting the position information of the first picture feature vector and the plurality of sub-pictures in the picture to be processed into a preset decoder to extract the feature vector, so as to obtain a second picture feature vector corresponding to the residual sub-picture of the plurality of sub-pictures after the target sub-picture is removed.
Specifically, the preset decoder and the preset encoder have the same structure, in the embodiment of the invention, 6 preset decoders are connected in series from beginning to end, the position information of the first picture feature vector and the plurality of sub-pictures in the to-be-processed picture is respectively input into a first preset decoder to perform feature extraction, the output vector of the first preset decoder is obtained, the output vector of the first preset decoder is input into a second preset decoder to perform feature extraction, the output vector of the last decoder is used as the input vector of the next decoder, the output vector of the last decoder is determined as the second picture feature vector corresponding to the rest of sub-pictures in the plurality of sub-pictures, and finally, a third picture special diagnosis vector corresponding to the to-be-processed picture is determined based on the first picture feature vector and the second picture feature vector.
207. And determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
For the embodiment of the present invention, in order to determine the third picture feature vector corresponding to the to-be-processed picture, step 207 specifically includes: determining a first weight coefficient corresponding to the first picture feature vector and a second weight coefficient corresponding to the second picture feature vector; and adding the first picture feature vector and the second picture feature vector based on the first weight coefficient and the second weight coefficient to obtain a third picture feature vector corresponding to the picture to be processed.
Specifically, first, a first weight coefficient corresponding to the first picture feature vector and a second weight coefficient corresponding to the second picture feature vector are respectively determined, the first weight coefficient and the first picture feature vector are multiplied to obtain a first product, meanwhile, the second weight coefficient and the second picture feature vector are multiplied to obtain a second product, and finally, the first product and the second product are added to obtain a third picture feature vector corresponding to the picture to be processed.
According to the other picture feature extraction method provided by the invention, compared with the mode that a feature extraction model is constructed by using clear and complete pictures at present and picture feature extraction is carried out based on the constructed feature extraction model, the picture to be processed in an actual service scene is obtained; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed; determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, thereby determining a clear target sub-picture in the picture to be processed by dividing the picture to be processed, extracting the first picture feature vector corresponding to the target sub-picture, determining second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures based on the first picture feature vector and the position information of the plurality of sub-pictures in the picture to be processed respectively, and finally determining the third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector, so that the situation that feature extraction is performed on an unclear picture by using a feature extraction model constructed by clear and complete pictures can be avoided, and the precision of picture feature extraction is improved.
Further, as a specific implementation of fig. 1, an embodiment of the present invention provides a device for extracting a picture feature, as shown in fig. 3, where the device includes: an acquisition unit 31, a division unit 32, an extraction unit 33, a first determination unit 34, and a second determination unit 35.
The obtaining unit 31 may be configured to obtain a picture to be processed in an actual service scene.
The dividing unit 32 may be configured to divide the to-be-processed picture into a plurality of sub-pictures, and determine a clear target sub-picture from the plurality of sub-pictures, where the clear target sub-picture is a picture that is not damaged, is not mosaic, and has a pixel value greater than a preset pixel value.
The extracting unit 33 may be configured to input the target sub-picture into a preset picture feature extraction model to perform feature extraction, so as to obtain a first picture feature vector corresponding to the target sub-picture.
The first determining unit 34 may be configured to determine, based on the first picture feature vector and the position information of the plurality of sub-pictures in the to-be-processed picture, a second picture feature vector corresponding to a remaining sub-picture of the plurality of sub-pictures from which the target sub-picture is removed.
The second determining unit 35 may be configured to determine a third picture feature vector corresponding to the to-be-processed picture based on the first picture feature vector and the second picture feature vector.
In a specific application scenario, in order to input the target sub-picture into a preset picture feature extraction model for feature extraction, a first picture feature vector corresponding to the target sub-picture is obtained, as shown in fig. 4, the extraction unit 33 includes a determining module 331, a splicing module 332, and an extraction module 333.
The determining module 331 may be configured to determine a pixel matrix corresponding to the target sub-picture.
The stitching module 332 may be configured to perform lateral stitching on each row of pixels in the pixel matrix, so as to obtain a fourth image feature vector corresponding to the target sub-image.
The extracting module 333 may be configured to input the fourth image feature vector into the preset image feature extraction model to perform feature extraction, so as to obtain a first image feature vector corresponding to the target sub-image.
In a specific application scenario, in order to input the fourth image feature vector into the preset image feature extraction model to perform feature extraction, a first image feature vector corresponding to the target sub-image is obtained, and the extraction module 333 includes an extraction sub-module, a summation sub-module, and an addition sub-module.
The extraction sub-module may be configured to input the fourth image feature vector to different attention subspaces in the attention layer to perform feature extraction, so as to obtain a fifth image feature vector of the target sub-image in the different attention subspaces.
The summation sub-module may be configured to multiply and sum a fifth picture feature vector of the target sub-picture in the different attention subspace with weights corresponding to the different attention subspaces, to obtain an attention layer output vector corresponding to the target sub-picture.
The adding sub-module may be configured to add the attention layer output vector and the fourth picture feature vector to obtain a sixth picture feature vector corresponding to the target sub-picture.
The extraction sub-module may be specifically configured to input the sixth picture feature vector into the feedforward neural network layer to perform feature extraction, so as to obtain a first picture feature vector corresponding to the target sub-picture.
In a specific application scenario, in order to train and construct the preset picture feature extraction model, the apparatus further comprises a construction unit 36.
The obtaining unit 31 may be further configured to obtain a sample picture in an actual service scene and an actual sample picture feature vector corresponding to the sample picture.
The dividing unit 32 may be further configured to divide the sample picture into a plurality of sample sub-pictures, and determine a clear target sample sub-picture from the plurality of sample sub-pictures.
The extracting unit 33 may be further configured to input the target sample sub-picture into an initial picture feature extraction model to perform feature extraction, so as to obtain a first sample picture feature vector corresponding to the target sample sub-picture.
The second determining unit 35 may be further configured to determine a second sample picture feature vector corresponding to the sample picture based on the first sample picture feature vector and position information of the plurality of sample sub-pictures in the sample picture, respectively.
The construction unit 36 may be configured to construct a loss function corresponding to the initial picture feature extraction model based on the actual sample picture feature vector corresponding to the sample picture and the second sample picture feature vector.
The construction unit 36 may be specifically configured to train the initial picture feature extraction model based on the loss function, and construct the preset picture feature extraction model.
In a specific application scenario, in order to construct a loss function corresponding to the initial picture feature extraction model, the construction unit 36 includes a calculation module 361 and a construction module 362.
The calculating module 361 may be configured to calculate respective vector differences at the same positions in the actual sample picture feature vector and the second sample picture feature vector.
The construction module 362 may be configured to construct a loss function corresponding to the initial image feature extraction model by calculating a sum of squares of the respective vector differences.
In a specific application scenario, in order to determine, based on the first picture feature vector and the position information of the plurality of sub-pictures in the to-be-processed picture, the second picture feature vector corresponding to the remaining sub-pictures in the plurality of sub-pictures, the first determining unit 34 may be specifically configured to input the first picture feature vector and the position information of the plurality of sub-pictures in the to-be-processed picture respectively into a preset decoder to extract the feature vector, so as to obtain the second picture feature vector corresponding to the remaining sub-pictures in the plurality of sub-pictures.
In a specific application scenario, in order to determine a third picture feature vector corresponding to the to-be-processed picture based on the first picture feature vector and the second picture feature vector, the second determining unit 35 may be specifically configured to determine a first weight coefficient corresponding to the first picture feature vector and a second weight coefficient corresponding to the second picture feature vector; and adding the first picture feature vector and the second picture feature vector based on the first weight coefficient and the second weight coefficient to obtain a third picture feature vector corresponding to the picture to be processed.
It should be noted that, for other corresponding descriptions of each functional module related to the image feature extraction device provided by the embodiment of the present invention, reference may be made to corresponding descriptions of the method shown in fig. 1, which are not repeated herein.
Based on the above method as shown in fig. 1, correspondingly, the embodiment of the present invention further provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, implements the following steps: acquiring a picture to be processed in an actual service scene; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed; and determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
Based on the embodiment of the method shown in fig. 1 and the device shown in fig. 3, the embodiment of the invention further provides a physical structure diagram of a computer device, as shown in fig. 5, where the computer device includes: a processor 41, a memory 42, and a computer program stored on the memory 42 and executable on the processor, wherein the memory 42 and the processor 41 are both arranged on a bus 43, the processor 41 performing the following steps when said program is executed: acquiring a picture to be processed in an actual service scene; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed; and determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector.
According to the technical scheme, the picture to be processed in the actual service scene is acquired; dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value; inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture; determining second picture feature vectors corresponding to the rest sub-pictures in the plurality of sub-pictures based on the first picture feature vectors and the position information of the plurality of sub-pictures in the to-be-processed picture respectively; determining a third picture feature vector corresponding to the to-be-processed picture based on the first picture feature vector and the second picture feature vector, determining a clear target sub-picture in the to-be-processed picture by dividing the to-be-processed picture, extracting the first picture feature vector corresponding to the target sub-picture, determining second picture feature vectors corresponding to the remaining sub-pictures after the target sub-picture is removed in the plurality of sub-pictures based on the first picture feature vector and the position information of the plurality of sub-pictures in the to-be-processed picture respectively, and finally determining the third picture feature vector corresponding to the to-be-processed picture based on the first picture feature vector and the second picture feature vector, so that the situation that feature extraction is performed on an unclear picture by using a feature extraction model constructed by clear and complete pictures can be avoided, and the precision of picture feature extraction is improved.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented in a general purpose computing device, they may be concentrated on a single computing device, or distributed across a network of computing devices, they may alternatively be implemented in program code executable by computing devices, so that they may be stored in a memory device for execution by computing devices, and in some cases, the steps shown or described may be performed in a different order than that shown or described, or they may be separately fabricated into individual integrated circuit modules, or multiple modules or steps within them may be fabricated into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. A picture feature extraction method, comprising:
acquiring a picture to be processed in an actual service scene;
Dividing the picture to be processed into a plurality of sub-pictures, and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value;
Inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture;
determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the picture to be processed;
Determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector;
The determining, based on the first picture feature vector and the second picture feature vector, a third picture feature vector corresponding to the to-be-processed picture includes:
determining a first weight coefficient corresponding to the first picture feature vector and a second weight coefficient corresponding to the second picture feature vector;
And adding the first picture feature vector and the second picture feature vector based on the first weight coefficient and the second weight coefficient to obtain a third picture feature vector corresponding to the picture to be processed.
2. The method of claim 1, wherein the inputting the target sub-picture into a preset picture feature extraction model for feature extraction to obtain a first picture feature vector corresponding to the target sub-picture comprises:
determining a pixel matrix corresponding to the target sub-picture;
Transversely splicing all rows of pixels in the pixel matrix to obtain a fourth picture feature vector corresponding to the target sub-picture;
And inputting the fourth picture feature vector into the preset picture feature extraction model to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture.
3. The method according to claim 2, wherein the preset picture feature extraction model is a preset encoder, the preset encoder includes an attention layer and a feedforward neural network layer, the inputting the fourth picture feature vector into the preset picture feature extraction model for feature extraction, and obtaining a first picture feature vector corresponding to the target sub-picture includes:
Inputting the fourth picture feature vector into different attention subspaces in the attention layer for feature extraction to obtain a fifth picture feature vector of the target sub-picture in the different attention subspaces;
multiplying and summing a fifth picture feature vector of the target sub-picture in the different attention subspaces with weights corresponding to the different attention subspaces to obtain an attention layer output vector corresponding to the target sub-picture;
Adding the attention layer output vector and the fourth picture feature vector to obtain a sixth picture feature vector corresponding to the target sub-picture;
and inputting the sixth picture feature vector into the feedforward neural network layer to perform feature extraction to obtain a first picture feature vector corresponding to the target sub-picture.
4. The method according to claim 1, wherein before the inputting the target sub-picture into a preset picture feature extraction model for feature extraction, the method further comprises:
Acquiring a sample picture in an actual service scene and an actual sample picture feature vector corresponding to the sample picture;
dividing the sample picture into a plurality of sample sub-pictures, and determining a clear target sample sub-picture from the plurality of sample sub-pictures;
inputting the target sample sub-picture into an initial picture feature extraction model to perform feature extraction to obtain a first sample picture feature vector corresponding to the target sample sub-picture;
determining a second sample picture feature vector corresponding to the sample picture based on the first sample picture feature vector and the position information of the plurality of sample sub-pictures in the sample picture respectively;
constructing a loss function corresponding to the initial picture feature extraction model based on the actual sample picture feature vector corresponding to the sample picture and the second sample picture feature vector;
Training the initial picture feature extraction model based on the loss function, and constructing the preset picture feature extraction model.
5. The method of claim 4, wherein constructing the loss function corresponding to the initial picture feature extraction model based on the actual sample picture feature vector corresponding to the sample picture and the second sample picture feature vector comprises:
Calculating each vector difference at the same position in the actual sample picture feature vector and the second sample picture feature vector;
and constructing a loss function corresponding to the initial picture feature extraction model by calculating the square sum of the vector differences.
6. The method according to claim 1, wherein the determining, based on the first picture feature vector and the position information of the plurality of sub-pictures in the to-be-processed picture, the second picture feature vector corresponding to the remaining sub-pictures of the plurality of sub-pictures from which the target sub-picture is removed includes:
And inputting the position information of the first picture feature vector and the plurality of sub-pictures in the picture to be processed into a preset decoder to extract the feature vector, so as to obtain a second picture feature vector corresponding to the residual sub-picture of the plurality of sub-pictures after the target sub-picture is removed.
7. A picture feature extraction apparatus, comprising:
the acquisition unit is used for acquiring the picture to be processed in the actual service scene;
the dividing unit is used for dividing the picture to be processed into a plurality of sub-pictures and determining a clear target sub-picture from the plurality of sub-pictures, wherein the clear target sub-picture is a picture which is not damaged, is not mosaic and has a pixel value larger than a preset pixel value;
the extraction unit is used for inputting the target sub-picture into a preset picture feature extraction model to perform feature extraction, so as to obtain a first picture feature vector corresponding to the target sub-picture;
the first determining unit is used for determining second picture feature vectors corresponding to the remaining sub-pictures of the plurality of sub-pictures after the target sub-picture is removed based on the first picture feature vectors and the position information of the plurality of sub-pictures in the to-be-processed picture respectively;
The second determining unit is used for determining a third picture feature vector corresponding to the picture to be processed based on the first picture feature vector and the second picture feature vector;
A second determining unit, configured to determine a first weight coefficient corresponding to the first picture feature vector and a second weight coefficient corresponding to the second picture feature vector; and adding the first picture feature vector and the second picture feature vector based on the first weight coefficient and the second weight coefficient to obtain a third picture feature vector corresponding to the picture to be processed.
8. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the computer program when executed by the processor implements the steps of the method according to any one of claims 1 to 6.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210026806.2A CN114387450B (en) | 2022-01-11 | 2022-01-11 | Picture feature extraction method and device, storage medium and computer equipment |
| PCT/CN2022/089692 WO2023134064A1 (en) | 2022-01-11 | 2022-04-27 | Image feature extraction method, apparatus, storage medium, and computer device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210026806.2A CN114387450B (en) | 2022-01-11 | 2022-01-11 | Picture feature extraction method and device, storage medium and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114387450A CN114387450A (en) | 2022-04-22 |
| CN114387450B true CN114387450B (en) | 2024-10-15 |
Family
ID=81201321
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210026806.2A Active CN114387450B (en) | 2022-01-11 | 2022-01-11 | Picture feature extraction method and device, storage medium and computer equipment |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN114387450B (en) |
| WO (1) | WO2023134064A1 (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114387450B (en) * | 2022-01-11 | 2024-10-15 | 平安科技（深圳）有限公司 | Picture feature extraction method and device, storage medium and computer equipment |
| CN115278293A (en) * | 2022-06-15 | 2022-11-01 | 平安科技（深圳）有限公司 | Virtual anchor generation method and device, storage medium and computer equipment |
| CN116484224A (en) * | 2023-04-23 | 2023-07-25 | 平安科技（深圳）有限公司 | A training method, device, medium and equipment for a multimodal pre-training model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102609690A (en) * | 2012-02-09 | 2012-07-25 | 北京海和鑫生信息科学研究所有限公司 | Method for evaluating quality of collected lower-half palm prints of living person |
| CN113869048A (en) * | 2021-09-30 | 2021-12-31 | 广州华多网络科技有限公司 | Commodity object search method and its device, equipment, medium and product |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005346287A (en) * | 2004-06-01 | 2005-12-15 | Denso Corp | Image recognizing method and apparatus |
| US8306366B2 (en) * | 2007-08-23 | 2012-11-06 | Samsung Electronics Co., Ltd. | Method and apparatus for extracting feature points from digital image |
| CN102521592B (en) * | 2011-11-30 | 2013-06-12 | 苏州大学 | Multi-feature fusion salient region extracting method based on non-clear region inhibition |
| CN107451454B (en) * | 2017-07-29 | 2020-03-24 | Oppo广东移动通信有限公司 | Unlocking control method and related product |
| CN112101395A (en) * | 2019-06-18 | 2020-12-18 | 上海高德威智能交通系统有限公司 | Image identification method and device |
| CN110598717B (en) * | 2019-09-12 | 2022-06-21 | 北京迈格威科技有限公司 | Image feature extraction method, device and electronic device |
| CN113223009B (en) * | 2021-04-16 | 2023-10-17 | 北京戴纳实验科技有限公司 | Clothes-wearing detection system |
| CN113626444B (en) * | 2021-08-26 | 2023-11-28 | 平安国际智慧城市科技股份有限公司 | Table query method, device, equipment and medium based on bitmap algorithm |
| CN114387450B (en) * | 2022-01-11 | 2024-10-15 | 平安科技（深圳）有限公司 | Picture feature extraction method and device, storage medium and computer equipment |
-
2022
- 2022-01-11 CN CN202210026806.2A patent/CN114387450B/en active Active
- 2022-04-27 WO PCT/CN2022/089692 patent/WO2023134064A1/en not_active Ceased
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102609690A (en) * | 2012-02-09 | 2012-07-25 | 北京海和鑫生信息科学研究所有限公司 | Method for evaluating quality of collected lower-half palm prints of living person |
| CN113869048A (en) * | 2021-09-30 | 2021-12-31 | 广州华多网络科技有限公司 | Commodity object search method and its device, equipment, medium and product |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114387450A (en) | 2022-04-22 |
| WO2023134064A1 (en) | 2023-07-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114387450B (en) | Picture feature extraction method and device, storage medium and computer equipment | |
| KR102191351B1 (en) | Method for semantic segmentation based on knowledge distillation | |
| CN109410261B (en) | Monocular image depth estimation method based on pyramid pooling module | |
| CN110222598B (en) | Video behavior identification method and device, storage medium and server | |
| CN109840477B (en) | Method and device for recognizing shielded face based on feature transformation | |
| CN111612024B (en) | Feature extraction method, device, electronic equipment and computer readable storage medium | |
| Rajapakse et al. | Color channel encoding with NMF for face recognition | |
| KR20190055447A (en) | Apparatus and method for generating and using neural network model applying accelerated computation | |
| CN109948441B (en) | Model training method, image processing method, device, electronic equipment and computer readable storage medium | |
| CN105976395B (en) | A kind of video target tracking method based on rarefaction representation | |
| CN111046738B (en) | Precision improvement method of light u-net for finger vein segmentation | |
| CN105868797A (en) | Network parameter training method, scene type identification method and devices | |
| CN111860672B (en) | Fine-grained image classification method based on block convolutional neural network | |
| CN113052868A (en) | Cutout model training and image cutout method and device | |
| CN111860276A (en) | Human body key point detection method, device, network device and storage medium | |
| CN113963009B (en) | Local self-attention image processing method and system based on deformable block | |
| CN113421258A (en) | Automatic cloud detection method based on high-resolution remote sensing image | |
| CN109992690B (en) | Image retrieval method and system | |
| CN114005046A (en) | Remote sensing scene classification method based on Gabor filter and covariance pooling | |
| CN114241585A (en) | Cross-age face recognition model training method, recognition method and device | |
| CN114897711B (en) | Image processing method, device, equipment and storage medium in video | |
| CN114723883B (en) | A three-dimensional scene reconstruction method, device, equipment and storage medium | |
| CN117542045B (en) | Food identification method and system based on space-guided self-attention | |
| CN115909332A (en) | Agricultural remote sensing image semantic segmentation method based on channel feature enhancement | |
| CN114926753A (en) | Method for rapidly extracting target scene information under condition of mass images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |