CN100462978C - An image retrieval method and system - Google Patents

Abstract

An image indexing system is prepared as using the first similarity set to compare image requiring to be indexed with each standard sample image in standard sample image set, indexing image to be indexed in image library according to the first similarity set and similarity set library and forming similarity set library by each image to be indexed in image library and the second similarity set of each standard sample image in standard sample image set.

Description

An image retrieval method and system

technical field

The invention relates to the field of data retrieval, in particular to an image retrieval method and system.

Background technique

Research on image retrieval has started since the 1970s. One is the image retrieval technology based on image text indexing (TBIR, Text-based Image Retrieval), which is also the most widely used and mature image retrieval technology. The retrieval method mainly uses the image text indexing method to describe the characteristics of the image, such as image description, source, etc. The other is content-based image retrieval (CBIR, Content-based Image Retrieval) technology, which represents the future development direction of image retrieval technology. Retrieval by Image Retrieval Techniques. With the rapid development of computer technology and communication technology, retrieving and browsing massive multimedia data including digital images has become an increasingly urgent problem.

The inventor noticed in the process of invention that for most of the image retrieval technologies based on image text indexing, there are the following deficiencies:

1) There is no objective and unified standard for text indexing itself, and it has a certain degree of subjectivity. For example, different "persons" may index different text information for the same image;

2) Through image text indexing information retrieval, the retrieval may fail due to the inconsistency between the retrieved text keywords and the image text index in the retrieved image database, and there may be a large number of irrelevant retrieval results in the retrieval results. That is, since the relationship between the tag and the image content is artificially set, there is no necessary correlation between the tag and the image content, and the real content of the image cannot be retrieved.

For the current content-based image retrieval technology, generally collect and process image resources, extract features such as color features, texture features, shape features, etc. Records that meet the threshold are output as a result in descending order of similarity. But the inventor has noticed that it has the following deficiencies:

1) Since the method of using the color histogram to calculate the similarity between images is relatively simple, it cannot reflect the spatial characteristics of the objects in the image;

2) Due to the difficulty of texture description, the retrieval of textures adopts the example query method, multiple interactions, and gradual refinement, resulting in slow query speed;

3) When searching based on shape features, the user can detect images with similar shapes from the image library by outlining the shape or outline of the image. There are two retrieval methods based on this feature: i) After the segmented image is extracted through edge extraction, the contour line of the target image is obtained, and the shape feature retrieval is carried out for this contour line. ii) Find the appropriate vector features directly for the graph to be used in the retrieval algorithm. Obviously, it is more complicated to deal with this kind of structured retrieval, more preprocessing and so on are required, which makes this method not only slow but also difficult to implement.

It can be seen from the above that the above technologies exist: 1) the query results returned by the retrieval are not accurate enough; or 2) the calculation is complex, requiring multiple interactions, and the query speed is slow. As a result, the existing image retrieval technology has not been widely used in practice.

Contents of the invention

The present invention provides an image retrieval method, system and device, which are used to solve the problems in the prior art that the retrieval efficiency and accurate retrieval results cannot be satisfied at the same time during image retrieval.

The invention provides an image retrieval method, comprising the following steps:

Comparing the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set;

According to the degree of similarity between the second similarity set in the similarity set library and the first similarity set, retrieve the image to be retrieved corresponding to the second similarity set from the image library, and the similarity set library It is a similarity set library composed of each image to be retrieved in the image library and the second similarity set of each standard sample image in the standard sample image set.

Preferably, further comprising the following steps:

When an image to be retrieved is added to the image database, a second similarity set between the added image to be retrieved and each standard sample image in the standard sample image set is added to the similarity set library.

Preferably, further comprising the following steps:

An image with a small similarity between images is used as a standard sample image in the standard sample image set.

Preferably, the similarity is compared according to feature vectors between images.

The present invention also provides an image retrieval system, comprising:

The first comparison module is used to compare the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set;

The second comparison module is used to compare the second similarity set between each image to be retrieved in the image library and each standard sample image in the standard sample image set;

A storage module, connected to the second comparison module, for storing a similarity set library composed of the second similarity sets compared by the second comparison module;

The retrieval module is connected with the first comparison module and the storage module, and is used for comparing the first similarity set compared with the first comparison module with the second similarity set in the similarity set library in the storage module The degree of similarity is high or low, and the image to be retrieved corresponding to the second similarity set is retrieved from the image database.

Preferably, it further includes:

The output module is connected with the retrieval module, and is used for outputting the image to be retrieved from the image database retrieved by the retrieval module.

Preferably, it further includes:

The adding module is connected with the second comparison module and the storage module, and is used to compare the added image to be retrieved by the second comparison module with each of the standard sample image sets when adding the image to be retrieved to the image library The second similarity set of the standard sample image is added to the similarity set library of the storage module.

Preferably, it further includes:

The standard sample image selection module is configured to use images with a small similarity between images as standard sample images in the standard sample image set.

Preferably, it further includes:

Feature vector input module, for the feature vector of input image;

The system retrieves images based on the similarity between feature vectors of the images.

The beneficial effects of the present invention are as follows:

Because the present invention first compares the second similarity set of each standard sample image between each image to be retrieved in the image library and the standard sample image set; and the similarity set library formed by the second similarity set compared; A standard sample image set determined in advance is used as the image feature measurement scale, so it has strong objectivity and consistency;

After the similarity set library is established, by comparing the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set; Retrieve an image. Therefore, by introducing the standard sample image set as the intermediary of comparison, the feature space of the image is expanded, more image feature information is included, and the accuracy of the retrieval result set is improved. At the same time, it is not necessary to compare the retrieved images with the images in the image library to calculate their similarity, but to use the standard sample image set as the intermediary for comparison, that is, to calculate in advance the differences between the images in the image library and the standard sample image set. The similarity of the standard sample image is used as its index vector, which improves the retrieval speed.

Further, since the method of determining the feature vector of the image is used to retrieve the similarity, the invention is simple, efficient and easy to implement.

Description of drawings

Fig. 1 is a schematic diagram of the implementation flow of the image retrieval method described in the embodiment of the present invention;

Fig. 2 is a schematic structural diagram of the image retrieval system described in the embodiment of the present invention;

Fig. 3 is a schematic diagram of the implementation of the image retrieval process described in the embodiment of the present invention.

Detailed ways

Specific embodiments of the present invention will be described below in conjunction with the accompanying drawings.

The purpose of the present invention is to propose a retrieval method that can improve the retrieval performance in view of the fact that the usual image retrieval method is complex in the retrieval process, the retrieval speed is slow, and the accuracy of the returned result set is not good, and a corresponding retrieval method is proposed accordingly Implementation. The implementation is simple, which not only improves the retrieval speed, but also improves the accuracy of the returned result set.

Figure 1 is a schematic diagram of the implementation process of the image retrieval method. As shown in the figure, the retrieval includes the following steps:

Step 101. Comparing each image to be retrieved in the image library with each standard sample image in the standard sample image set to obtain a second similarity set for each image to be retrieved;

In a preferred implementation, when selecting a standard sample image, the similarity between images can be used as a basis, and an image with a small similarity between images is used as a standard sample image in the standard sample image set, so that the image in the similarity set library The eigenvector of the image can more accurately describe the characteristics of the image. In other words, the eigenvector of the image in the similarity set library can better express the mutual differences between the images in the similarity set library, so that the subsequent retrieval accuracy Becomes high.

Step 102, a similarity set library composed of each second similarity set;

Through the preparation of the above two steps, the similarity set library of the image to be retrieved is established, and the subsequent retrieval process can be changed from the previous retrieval of image content to similarity retrieval, thereby avoiding the disadvantages of direct retrieval based on image content .

In implementation, if necessary, when adding the image to be retrieved to the image database, the second similarity set between the added image to be retrieved and each standard sample image in the standard sample image set may also be added to the similarity set library.

Step 103, comparing the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set;

Step 104: Retrieve the image to be retrieved from the image library according to the first similarity set and the similarity set library.

In a preferred implementation, according to the degree of similarity between the second similarity set in the similarity set library and the first similarity set, the image to be retrieved corresponding to the second similarity set is retrieved from the image library, Since the images to be retrieved through similarity retrieval are not exactly matched one by one, when outputting the retrieval results, they can be output according to the level of similarity matching according to actual needs.

In implementation, the similarity can be obtained according to the matching level of feature vectors between images, and the number of image feature vectors can be selected as required, and more feature vectors can be selected if high precision is required.

Next, an example is used to illustrate the implementation of the retrieval method based on image similarity comparison, including the following steps:

1) Determine the similarity set library formed by the similarity between each standard sample image in the standard sample image set and each standard sample image in the standard sample image set obtained by using the image feature vector.

In the embodiment, S indicates that the image is a standard sample image, [k] indicates the serial number of the standard sample image; D indicates that the image is an image to be retrieved in the image database, [n] indicates the serial number of the image to be retrieved, and IS[i][j] indicates the number of the image to be retrieved. The similarity between the i standard sample image and the jth image to be retrieved, DX[i]=(IS[i][1], IS[i][2], ..., IS[i][K]) is the similarity set (second similarity set) of the i-th standard sample image.

Input standard sample images (S[1], S[2], ..., S[K]) and image library images (D[1], D[2], ..., D[n]), images The library images are compared with the standard sample images one by one, and the similarity IS[i][j] between them is calculated respectively, as shown in the following table:

Thus, the similar feature vector set DX[i]=(IS[i ][1], IS[i][2], ..., IS[i][K]).

2) After repeating the process 1) to calculate and save the feature vectors of all image library images relative to the standard sample images (S[1], S[2], ..., S[K]), a similarity set library is formed.

3) Determine the similarity set (first similarity set) by using the feature vector of the image to be retrieved.

Input the standard sample image (S[1], S[2], ..., S[K]) and the image to be retrieved KEY, the image to be retrieved is compared with the standard sample image for similarity, and the relationship between them is calculated respectively. The preliminary similarity KS[j], so, obtain the similarity set KEYX=of the similar feature vector of the image KEY to be retrieved relative to the standard sample image (S[1], S[2], . (KS[1], KS[2], ..., KS[K]).

4) Carry out image retrieval calculation.

Calculate the similarity FS[i] between the image feature vector KEYX to be retrieved and the image feature vector DX[i] in the image library similarity set as the final similarity. Assuming that the value range of preliminary similarity KS[j] and IS[i][j] is 0-100, the calculation method of similarity FS[i] can be defined as follows, FS[i]=100-(ABS(KS[ 1]-IS[i][1])+ABS(KS[2]-IS[i][2])+...+ABS(KS[K]-IS[i][K]))/K , where ABS(a) represents the absolute value calculation of a.

In the specific implementation, there are many implementations based on the principle of using the similarity set and the similarity set library obtained from the images to be retrieved. Therefore, this embodiment provides a calculation method for the similarity FS[i], but it is easy to know that it is not limited to this way.

5) Output query results.

FS[i] is sorted in reverse order, and the top result is the most similar retrieval output result.

In the implementation, before image retrieval, the similar feature vectors of all images in the image library relative to the standard sample images should be determined first, and the similar feature vectors of the image library images relative to the standard sample images and the similar feature vectors of the retrieved images relative to the standard sample images can be used The similarity between them gets the most similar retrieval output.

In order to enhance the retrieval effect, the following auxiliary measures can be taken in the preferred implementation:

1) Select a standard sample image set. When selecting standard sample images, images with small similarities between images are used as standard sample images in the standard sample image set, and those images with relatively large differences between each image are selected as standard sample images as much as possible. For example, one can be selected as a sample image from images of different categories such as person images and landscape images.

2) Select a preliminary similarity calculation method between images. When choosing a preliminary similarity calculation method, you can choose those calculation methods that can reflect the differences between images and have strong objectivity and consistency.

The present invention also provides an image retrieval system. FIG. 2 is a schematic structural diagram of the image retrieval system. As shown in the figure, the image retrieval system includes: a first comparison module, a second comparison module, a storage module, and a retrieval module, wherein:

The first comparison module is used to compare the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set;

The second comparison module is used to compare the second similarity set between each image to be retrieved in the image database and each standard sample image in the standard sample image set;

The storage module is connected to the second comparison module, and is used to store the similarity set library composed of the second similarity sets compared by the second comparison module;

The retrieval module is connected with the first comparison module and the storage module, and is used for retrieving from the image library to be retrieved according to the first similarity set compared by the first comparison module and the similarity set library in the storage module. image.

In the preferred implementation, it can further include:

The output module is connected with the retrieval module, and is used for outputting the image to be retrieved corresponding to the second similarity set retrieved by the retrieval module from the image database.

The adding module is connected with the second comparison module and the storage module, and is used to compare the added image to be retrieved by the second comparison module with each of the standard sample image sets when adding the image to be retrieved to the image library The second similarity set of the standard sample image is added to the similarity set library of the storage module.

The standard sample image selection module is configured to use images with a small similarity between images as standard sample images in the standard sample image set.

In a preferred implementation, it may further include:

The feature vector input module is used to input the feature vectors of the images; the image retrieval system retrieves the images according to the similarity between the feature vectors of the images.

Figure 3 is a schematic diagram of the implementation of the image retrieval process. Briefly, as shown in the figure, the image retrieval can be implemented according to the following basic processes:

Step 301, adding images to be retrieved in the image library;

Step 302, calculating and adding feature vectors to be retrieved;

Step 303, the feature vector of the image to be retrieved in the image library;

Step 304, input the image to be retrieved;

Step 305, calculating the feature vector of the image to be retrieved;

Step 306, using the image feature vector of the image library to perform image retrieval calculation;

Step 307, output the query result.

The above implementation will be described below in combination with examples.

Suppose now there is an image library with 1 million images.

First, 5 standard sample images (S[1], S[2], ..., S[5]) are selected as the standard sample image set.

Secondly, calculate the similar feature vector DX[i]=(IS[i][1],IS[i][2],...,IS[ i][5]), i=1, 2, . . . , 1000000.

Again, calculate the similar feature vector KEYX=(KS[1], KS[2], . . . , KS[5]) of the image KEY to be retrieved relative to the five standard sample images.

Then, calculate the final similarity FS[i] between the feature vector KEYX of the image to be retrieved and the feature vector DX[i] of the image database, i=1, 2, . . . , 1000000.

Finally, FS[i] (i=1, 2, . . . , 1000000) is sorted in reverse order, and the top results (for example, the top 10) are the most similar retrieval output results.

Through the solution described in the present invention, a faster and more accurate image retrieval effect can be obtained very effectively on the basis of the prior art. At the same time, in the retrieval scheme based on image similarity comparison provided by the present invention, different preliminary similarity calculation methods can be used to obtain the desired retrieval effect according to different requirements. Simple to implement and easy to operate.

As can be seen from the foregoing embodiments, the present invention: 1) proposes a new image feature extraction method, adopts a standard sample image set determined in advance as the image feature (vector) measurement scale, which has strong objectivity and consistency; 2 ) to determine the image features (vectors) is relatively simple. It is not necessary to compare the retrieved image with the images in the image library to calculate their similarity one by one, but to introduce a standard sample image set as a comparison intermediary, that is, to calculate the image library in advance The similarity between the image in the image and each standard sample image in the standard sample image set is used as its index vector, which improves the retrieval speed; 3) Due to the introduction of the standard sample image set as an intermediary for comparison, the feature space of the image is expanded, including more The feature information of the map improves the accuracy of the retrieval result set.

It can also be seen from the above implementation that, due to the ubiquity of various existing image retrieval methods, the accuracy and speed of the returned result sets of large-scale image database retrieval are not high, and it is difficult to meet the actual application requirements. Therefore, the present invention proposes a retrieval scheme based on image similarity comparison on the basis of existing image retrieval methods. By adopting the scheme of the present invention, the accuracy of the returned result set of image retrieval can be effectively improved, and at the same time, the image retrieval speed can be increased. In addition, it can also be applied to automatic matching of image similarities between two or more image databases for more retrieval applications.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims (9)

1. an image retrieval method, is characterized in that, comprises the steps: Comparing the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set; According to the degree of similarity between the second similarity set in the similarity set library and the first similarity set, retrieve the image to be retrieved corresponding to the second similarity set from the image library, and the similarity set library It is a similarity set library composed of each image to be retrieved in the image library and the second similarity set of each standard sample image in the standard sample image set. 2. The method according to claim 1, further comprising the steps of: When an image to be retrieved is added to the image database, a second similarity set between the added image to be retrieved and each standard sample image in the standard sample image set is added to the similarity set library. 3. The method of claim 1, further comprising the steps of: An image with a small similarity between images is used as a standard sample image in the standard sample image set. 4. The method according to any one of claims 1 to 3, wherein the similarity is compared according to feature vectors between images. 5. An image retrieval system, characterized in that, comprising: The first comparison module is used to compare the first similarity set of each standard sample image between the image to be retrieved and the standard sample image set; The second comparison module is used to compare the second similarity set between each image to be retrieved in the image library and each standard sample image in the standard sample image set; A storage module, connected to the second comparison module, for storing a similarity set library composed of the second similarity sets compared by the second comparison module; The retrieval module is connected with the first comparison module and the storage module, and is used for comparing the first similarity set compared with the first comparison module with the second similarity set in the similarity set library in the storage module The degree of similarity is high or low, and the image to be retrieved corresponding to the second similarity set is retrieved from the image database. 6. The system of claim 5, further comprising: The output module is connected with the retrieval module, and is used for outputting the image to be retrieved from the image database retrieved by the retrieval module. 7. The system of claim 5, further comprising: The adding module is connected with the second comparison module and the storage module, and is used to compare the added image to be retrieved by the second comparison module with each of the standard sample image sets when adding the image to be retrieved to the image library The second similarity set of the standard sample image is added to the similarity set library of the storage module. 8. The system of claim 5, further comprising: The standard sample image selection module is configured to use images with a small similarity between images as standard sample images in the standard sample image set. 9. The system according to any one of claims 5 to 8, further comprising: Feature vector input module, for the feature vector of input image; The system retrieves images based on the similarity between feature vectors of the images.

Images