CN102306179B - Image content retrieval method based on hierarchical color distribution descriptor - Google Patents
Image content retrieval method based on hierarchical color distribution descriptor Download PDFInfo
- Publication number
- CN102306179B CN102306179B CN 201110249085 CN201110249085A CN102306179B CN 102306179 B CN102306179 B CN 102306179B CN 201110249085 CN201110249085 CN 201110249085 CN 201110249085 A CN201110249085 A CN 201110249085A CN 102306179 B CN102306179 B CN 102306179B
- Authority
- CN
- China
- Prior art keywords
- image
- value
- minimum
- maximum
- retrieval
- 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
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000001914 filtration Methods 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000006835 compression Effects 0.000 claims description 5
- 238000007906 compression Methods 0.000 claims description 5
- 238000000926 separation method Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 description 7
- 230000008447 perception Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000004043 dyeing Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 210000001328 optic nerve Anatomy 0.000 description 1
- 229920000642 polymer Polymers 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000004753 textile Substances 0.000 description 1
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to an image content retrieval method based on a hierarchical color distribution descriptor, which comprises the following steps that: 1, a user inputs an image to be queried and a retrieval requirement into a retrieval system; 2, the retrieval system constructs the hierarchical color distribution descriptor of the image to be queried; 3, the retrieval system performs hierarchical filtering on a characteristic database by using the hierarchical color distribution descriptor of the image to be queried until the retrieval requirement is met to obtain a final characteristic database; 4, the retrieval system searches a corresponding image from an image database according to the final characteristic database and feeds the searched image back to the user; and 5, the user performs further searching according to a feedback result. By the image content retrieval method based on the hierarchical color distribution descriptor, the hierarchical color distribution descriptor is constructed to accurately describe the image, so that image content retrieval efficiency can be greatly improved.
Description
Technical Field
The invention relates to image retrieval, in particular to an image content retrieval method based on a hierarchical color distribution descriptor.
Background
With the rapid development of science and technology, the popularization of computers and broadband networks, the popularization of digital acquisition devices and digital storage devices, and the quantity of various multimedia information data including characters, images, videos, audios, and the like, are increasing at an incredible speed. How to find useful information from massive amounts of data as soon as possible has become a serious problem, and therefore, efficient access to data, particularly to visual information data, has become increasingly important.
In order to effectively solve the problem, a content-based retrieval method is proposed, which automatically acquires the content of the visual information by analyzing the characteristics of the visual information, thereby opening up a way for efficient information access. At present, in various industries such as medical treatment, textile printing and dyeing, weather analysis, library museums, security agencies, movie televisions and the like, some content-based information access systems, such as QBIC, photoblook, point, visual search and the like, have appeared. Content-based retrieval has become a specialized field of research, and its research results have contributed to the resolution of many problems, and more visual information retrieval systems are under development.
Currently, in content-based retrieval methods, color, texture, and spatial distribution are the most prominent features for performing the retrieval. Methods for describing the color, texture and spatial distribution of target information include a geometric parameter method, a moment invariant method, an accumulative histogram method, a color layout method and the like.
But in fact, the feature-based retrieval method is far from being solved well, and the key point is that the description of the color, texture and spatial distribution of the target information is a very complex problem, and often needs to be described by using a method including geometric, statistical or morphological methods, so that the description can be similar to human perception. Furthermore, the human perception of an image is not only a reflection result of the optic nerve, but also is closely related to the human knowledge about the real world, so that the human perception can only be approximated to a certain extent by mathematical language on the basis of experiments.
One prior art image retrieval method is based on invariant moment image retrieval method, moment is a statistical form of the image, its calculation uses all relevant pixel points in the image or region, for a digital image function f (x, y), if it is piecewise continuous and is not zero only at a limited point on the XY plane, it can be proved that its respective moment exists. The p + q moment of f (x, y) is defined as:
the p + q order central moment of f (x, y) is defined as:
wherein,
i.e. the coordinates of the center of mass.
The normalized central moment of f (x, y) can be expressed as:
where γ is (p + q)/2+1, and p + q is 2, 3, ….
The normalized second and third central moments are combined to yield 7 rotation and scale invariant moments φ for the image1，φ2，…，φ7And counting each image in the image library to obtain 7 invariant moments of each image as characteristic data, and comparing the characteristic data with the 7 invariant moments of the image to be inquired to judge the similarity of the images.
Although the feature data of the invariant moment method is invariant to the rotation and scaling of the image, since such statistic has no clear physical meaning, the image cannot be accurately described, and thus the retrieval result is quite rough and cannot meet the requirement of efficient data access.
Disclosure of Invention
The invention aims to provide an image content retrieval method based on a hierarchical color distribution descriptor, which constructs the hierarchical color distribution descriptor to accurately describe an image and realizes efficient image content retrieval.
In order to achieve the above object, the present invention provides an image content retrieval method based on a hierarchical color distribution descriptor, comprising the steps of: step 1, a user inputs an image to be inquired and a retrieval requirement to a retrieval system; step 2, the retrieval system constructs a hierarchical color distribution descriptor of the image to be queried; step 3, the retrieval system uses the hierarchical color distribution descriptor of the image to be queried, a hierarchical filtering feature database,obtaining a final characteristic database until the retrieval requirement is met; step 4, the retrieval system searches out a corresponding image from the image database according to the final feature database and feeds back the searched image to the user; step 5, the user further searches according to the feedback result; wherein, the step 2 specifically comprises the following steps: step 2.1, the retrieval system divides the image to be inquired into K1×K2Each image block and calculating an average value of each image block, wherein K1And K2Are all even numbers; when the image to be inquired is a gray image, the average value is the average value of all pixel gray values of the image block, and when the image to be inquired is a color image, the average value is the average value of all color component brightness values of the image block; step 2.2, construct the directional mean vector C, C ═ C1，c2，c3，....，cK1×K2}; step 2.3, compress the directional mean vector and extract the front K of the compressed vector3One component constitutes a hierarchical color distribution descriptor E, E ═ E of the image to be queried1，e2，e3，....，eK3In which K is3＜K1×K2。
The above-mentioned image content retrieval method based on a hierarchical color distribution descriptor, wherein,
n1＝2，3，......；
n2＝2，3，......；K3＝K1+K2。
in the image content retrieval method based on the hierarchical color distribution descriptor, if the average value of each image block of the image to be queried calculated in step 2.1 is equal, step 2.2 specifically includes making each component of the directional average vector C equal to the average value.
The above-mentioned based on hierarchical color distribution tracingThe image content retrieval method of the character, wherein if the average values of the image blocks of the image to be queried calculated in step 2.1 are not completely the same, step 2.2 specifically comprises the following steps: 2.2.1, selecting a pair of maximum-minimum value pairs with the minimum spacing distance from the average values of the same circle of image blocks according to the sequence from outside to inside; step 2.2.2, determining the rotation direction according to the maximum value-minimum value pair selected in the step 2.2.1; step 2.2.3, traversing all image blocks of the image to be queried according to the spiral rotation method from outside to inside from the maximum value to the minimum value of the maximum value-minimum value pair selected in step 2.2.1 to the minimum value in the rotation direction determined in step 2.2.2, and forming the average values of the image blocks into a directional average value vector C according to the traversal sequence, wherein C is { C ═ C { (C) } C { (C {1，c2，c3，....，cK1×K2}。
The image content retrieval method based on the hierarchical color distribution descriptor includes the following specific steps in step 2.2.1: firstly, selecting a pair of average value maximum value-minimum value pairs with minimum spacing distance from the edge image blocks of the image to be inquired, and if the average value maximum value-minimum value pairs cannot be selected from the edge image blocks of the image to be inquired, selecting from the inner ring according to the sequence from outside to inside; finding out all maximum values and minimum values from the average values of the same circle of image blocks, supposing that P maximum values and Q minimum values are found, forming P multiplied by Q maximum value-minimum value pairs, calculating the spacing distance S between each pair of maximum values and minimum values, and when calculating the spacing distance S, if the calculated value of the spacing distance S is larger than (2K)1+2 K2-4)/2, so that S ═ 2K1+2 K2-4) -S; if only one maximum-minimum value pair with the minimum spacing distance exists through calculation, the maximum-minimum value pair is the maximum-minimum value pair with the minimum spacing distance selected; if a plurality of maximum-minimum value pairs with the minimum spacing distance exist through calculation, a pair is determined by comparing adjacent items of the maximum value or the minimum value, and if the pair cannot be determined by comparing the adjacent items of the maximum value or the minimum value, the pair is selected randomly.
As described aboveThe image content retrieval method based on the hierarchical color distribution descriptor, wherein the step 2.2.2 specifically includes: if the maximum-minimum pair of values S selected in step 2.2.1 is less than (2K)1+2K24)/2, the direction of rotation is determined to be the direction from the maximum value to the minimum value over the minimum separation distance; if the spacing distance S of the maximum value-minimum value pair selected in the step 2.2.1 is equal to (4K-4)/2, determining the rotation direction through adjacent items of the maximum value or the minimum value; and if the spacing distance S of the maximum value-minimum value pair selected in the step 2.2.1 is equal to (4K-4)/2 and the average values of other image blocks in the same circle are equal except the maximum value and the minimum value, selecting a pair of maximum value-minimum value pairs from the inner circle of the circle according to the sequence from outside to inside to determine the rotating direction.
The above-mentioned image content retrieval method based on a hierarchical color distribution descriptor, wherein the max-min value pair is replaced by a max-next-max value pair.
The above image content retrieval method based on hierarchical color distribution descriptors, wherein the maximum-minimum value pairs are replaced by minimum-next-minimum value pairs.
The above-mentioned hierarchical color distribution descriptor-based image content retrieval method, wherein in said step 2.3 said directional mean vector C is compressed using a HAAR wavelet compression method.
The image content retrieval method based on the hierarchical color distribution descriptor described above, wherein the step 3 specifically includes the following steps: step 3.1, the retrieval system determines the stage number J of the hierarchical retrieval according to the retrieval requirement input in the step 1, wherein J is an integer and J is more than or equal to 1 and less than or equal to K3(ii) a Step 3.2, setting J to be 0, and setting J to be less than or equal to J; step 3.3, j equals j +1, the retrieval system combines the (j-1) filtered feature database with the jth component E of the vector EjPerforming feature matching, and filtering out the (j-1) filtered feature database and the component ejUnmatched feature data are obtained, and a feature database subjected to j times of filtering is obtained; step 3.4, judging whether J is smaller than J, if so, returning to the step 3.3, if not,it is indicated that the final feature database meeting the search requirements has been obtained.
The image content retrieval method based on the hierarchical color distribution descriptor has the following advantages:
the image content retrieval method based on the hierarchical color distribution descriptor uses the hierarchical color distribution descriptor to describe the image, and the hierarchical color distribution descriptor contains color information, space distribution information and texture information of the image and can accurately and visually describe the image; the hierarchical color distribution descriptor has the advantages of no deformation in rotation and scaling and good anti-interference capability; the hierarchical color distribution descriptor has a simple structure and small data volume, and can greatly reduce the calculated amount and improve the retrieval efficiency during retrieval;
the image content retrieval method based on the hierarchical color distribution descriptor performs feature matching by using the hierarchical color distribution descriptor, and can perform hierarchical retrieval due to the hierarchical characteristic of the hierarchical color distribution descriptor, thereby meeting the requirement of efficient access to visual information; and each time the feature matching only uses one feature, a large amount of feature data is filtered out every time the feature matching is carried out, so that the calculated amount is greatly reduced, and the retrieval speed is improved.
Drawings
The image content retrieval method based on the hierarchical color distribution descriptor of the present invention is given by the following embodiments and the accompanying drawings.
Fig. 1 is a flow chart of the image content retrieval method based on hierarchical color distribution descriptors of the present invention.
FIG. 2 is a diagram illustrating image partitioning according to an embodiment of the present invention.
FIG. 3 is a schematic illustration of determining a rotational direction in an embodiment of the present invention.
FIG. 4 is a schematic diagram of the construction of a directional mean vector in an embodiment of the present invention.
FIG. 5 is a flow chart of hierarchical feature matching in an embodiment of the present invention.
Detailed Description
The image content retrieval method based on the hierarchical color distribution descriptor according to the present invention will be described in further detail with reference to fig. 1 to 5.
Referring to fig. 1, the image content retrieval method based on hierarchical color distribution descriptor of the present invention includes the steps of:
step 2, the retrieval system constructs a hierarchical color distribution descriptor of the image to be queried;
step 2.1, the retrieval system divides the image to be inquired into K1×K2Each image block and calculating an average value of each image block, wherein K1And K2Are all even numbers;
preferably, the first and second liquid crystal films are made of a polymer,
n1＝2，3，......；
n2＝2，3，......；
when the image to be inquired is a gray image, the average value is the average value of all pixel gray values of the image block, and when the image to be inquired is a color image, the average value is the average value of all color component brightness values of the image block;
step 2.2, construct the directional mean vector C, C ═ C1，c2，c3，....，cK1×K2}；
Step 2.3, compress the directional mean vector and extract the front K of the compressed vector3One component constitutes a hierarchical color distribution descriptor E, E ═ E of the image to be queried1，e2，e3，....，eK3In which K is3＜K1×K2；
Preferably, K3＝K1+K2；
Step 3, the retrieval system uses the hierarchical color distribution descriptor of the image to be queried to filter the feature database in a hierarchical manner until the retrieval requirement is met, and a final feature database is obtained;
step 4, the retrieval system searches out a corresponding image from the image database according to the final feature database and feeds back the searched image to the user;
and 5, the user further searches according to the feedback result.
The hierarchical color distribution descriptor in the image content retrieval method based on the hierarchical color distribution descriptor comprises the color information, the space distribution information and the texture information of the image, has the invariance of rotation and scaling, can accurately describe the image, has the hierarchical characteristic, can carry out hierarchical retrieval by utilizing the hierarchical color distribution descriptor, and can improve the image retrieval efficiency.
The image content retrieval method based on the hierarchical color distribution descriptor according to the present invention will now be described in detail by specific embodiments:
the image content retrieval method based on the hierarchical color distribution descriptor of the embodiment comprises the following steps:
s100, inputting an image to be queried and a retrieval requirement to a retrieval system by a user;
for example, the size of the input image to be queried is M × N;
the retrieval requirement determines the number of levels of hierarchical retrieval;
step S200, the retrieval system constructs a hierarchical color distribution descriptor of the image to be queried;
step S210, the retrieval system divides the image to be inquired into K multiplied by K image blocks, wherein K is an even number, and the average value of each image block is calculated;
when the image to be inquired is a gray image, the average value is the average value of all pixel gray values of the image block, and when the image to be inquired is a color image, the average value is the average value of all color component brightness values of the image block;
in this embodiment, a uniform division method (i.e., K) is adopted for image division1＝K2K), but the invention is not limited thereto, and non-uniform partitioning methods, i.e. K, may also be employed1≠K2；
Preferably, the value range of K is 4, 8 and 16;
for example, taking K ═ 4, as shown in fig. 2, the retrieval system divides the image to be queried into 4 × 4 image blocks, and the 4 × 4 image blocks form a 4 × 4-order array;
arranging the average values in a row-by-row and column-by-column order to form an average value vector A, A ═ a1，a2，a3，....，ai，....，a16In which ai(i 1, 2.., 16) represents an average value of the image blocks, as shown in fig. 2;
the average value vector A contains color information and spatial distribution information of an image and has scaling invariance;
step S220, construct the directional mean vector C, C ═ C1，c2，c3，....，cK1×K2}；
Step S221, selecting a pair of maximum-minimum value pairs with minimum spacing distance from the average value of the same circle of image blocks according to the sequence from outside to inside;
should be selected from the edge image blocks (outermost circles) of the image to be queried first, specifically as follows:
taking the average value of the image blocks in the first row and the first column as a first item, and arranging the average value of the edge image blocks of the image to be inquired in a clockwise (or anticlockwise) direction to form a vector B;
finding all the maximum values and the minimum values from the vector B, and if P maximum values and Q minimum values are found, forming P multiplied by Q maximum value-minimum value pairs, calculating the spacing distance S between each pair of the maximum values and the minimum values, and if the calculated spacing distance S is larger than (4K-4)/2, making S (4K-4) -S;
if only one maximum-minimum value pair with the minimum spacing distance exists through calculation, the maximum-minimum value pair is the maximum-minimum value pair with the minimum spacing distance selected;
if, by calculation, there are a plurality of maximum-minimum value pairs with the smallest separation distance, then determining a pair by comparing adjacent terms of the maximum (or minimum), for example, comparing adjacent terms of the maximum, the maximum-minimum value pair which appears the larger value first among adjacent terms is determined as the maximum-minimum value pair to be selected; if a pair can not be determined by comparing adjacent items of the maximum value (or the minimum value), for example, adjacent items of each pair of the maximum value-minimum value pair are correspondingly equal, then a pair can be arbitrarily selected from the adjacent items;
if the average values of all the edge image blocks of the image to be inquired are equal, the maximum value-minimum value pair cannot be selected from the edge image blocks of the image to be inquired, at the moment, a pair is selected from the inner rings according to the sequence from outside to inside, namely, the pair is selected from the outermost ring, if the pair is not selected, the pair is selected from the first inner ring closest to the outermost ring, if the pair is not selected, the pair is selected from the second inner ring closest to the first inner ring, and the like;
for example, arranging the average value of the edge image blocks of the image to be queried in a clockwise direction to form a vector B ═ a1，a2，a3，a4，a8，a12，....，a5In vector B, there is only one maximum and one minimum, respectively a1And a12As shown in fig. 2;
step S222, determining the rotation direction according to the maximum value-minimum value pair selected in the step S221;
if the spacing distance S of the maximum value-minimum value pair selected in the step S221 is less than (4K-4)/2, the rotating direction is determined to be the direction from the maximum value to the minimum value through the minimum spacing distance;
if the spacing distance S of the maximum-minimum value pair selected in step S221 is equal to (4K-4)/2, the rotation direction is determined by the adjacent terms of the maximum value (or the minimum value), for example, the direction in which the larger value occurs first among the adjacent terms of the maximum value is taken as the rotation direction;
the rotation direction is either clockwise or counterclockwise;
if the spacing distance S of the maximum-minimum value pairs selected in step S221 is equal to (4K-4)/2, and the average values of other image blocks of the same circle are equal except for the maximum value and the minimum value, which indicates that the rotation direction cannot be determined by the maximum-minimum value pairs of the circle, at this time, a pair of maximum-minimum value pairs is selected from the inner circle of the circle in the sequence from outside to inside to determine the rotation direction, and the method for selecting a pair of maximum-minimum value pairs from the inner circle of the circle refers to step S221, and the method for determining the rotation direction refers to the above steps;
for example, the maximum value a1And a minimum value a12The spacing distance S is 4, less than (4K-4)/2 is 6, and the rotation direction is determined to be counterclockwise (from a)12Starting over a minimum separation distance to a1Direction) as shown in fig. 3;
step S223, traversing all image blocks of the image to be queried according to the spiral rotation method from outside to inside from the maximum value to the minimum value of the maximum-minimum value pair selected in step S221 to the minimum value in the rotation direction determined in step S222, and forming an oriented average vector C by the average values of the image blocks according to the traversal order, where C is { C ═1，c2，c3，....，cK×K}；
The oriented mean vector C has rotational invariance, which means that the vector C is invariant to rotation of integer multiples of 90 ° such as ± 90 °, ± 180 °, ± 270 °,. the like, and thus, the oriented mean vector C has rotational and scaling invariance and contains color information and spatial distribution information of an image;
in this embodiment, the rotation direction is determined by selecting a pair of maximum-minimum pairs of the average values, and an oriented average vector is constructed on the basis of the pair of maximum-minimum pairs of the average values, but the present invention is not limited thereto, and the rotation direction may also be determined by selecting a pair of maximum-minimum pairs of the average values, or minimum-minimum pairs of the average values, so as to construct an oriented average vector;
for example, the directional mean vector C ═ a1，a2，a3，a4，a8，a12，a16，a15，a14，a13，a9，a5，a7，a6，a11，a10The sequence of traversing all image blocks of the image to be queried is shown in fig. 4;
if the average value of each image block of the image to be queried calculated in step S210 is equal, making each component of the directional average vector C equal to the average value;
step S230, compressing the directional average value vector, and extracting the first K items of the compressed vector to form a hierarchical color distribution descriptor of the image to be inquired;
in this embodiment, the directional average vector C is compressed by using a HAAR wavelet compression method to obtain a compressed vector D, where D is { D ═ D }1，d2，d3，....，dK×KExtracting the first 2K items of the vector D to form a new vector E, and E ═ E { (E) }1，e2，e3，....，e2KThe vector E is the hierarchical color distribution descriptor of the image to be inquired;
in the vector E, E1Is the mean value of the image to be queried (i.e. the mean value of the entire image), e2～e2KThe high-order components of the image to be inquired can be used for describing the texture of the image;
because the vector E is obtained after the directional average value vector C is compressed, the rotation and scaling invariance of the vector is not changed by compression, the vector E has rotation and scaling invariance and good anti-interference capability, and simultaneously contains color information, space distribution information and texture information of an image, the description of the image is very intuitive and is similar to human feeling, namely the vector E can accurately describe the image;
each component in the vector E is data after compression processing, and the dimension of the vector E is 2K (much smaller than K multiplied by K), so that the vector E has a simple structure and small data volume, and the calculation amount can be greatly reduced and the retrieval efficiency can be improved by using the vector E for retrieval;
step S300, the retrieval system uses the hierarchical color distribution descriptor of the image to be queried to filter the feature database in a hierarchical manner until the retrieval requirement is met, and a final feature database is obtained;
the retrieval system is provided with an image database, hierarchical color distribution descriptors of each image in the image database are constructed, and a set of the hierarchical color distribution descriptors forms a feature database of the retrieval system;
referring to FIG. 5, in step S310, the retrieval system determines the number of stages J of the hierarchical retrieval according to the retrieval request input in step S100, wherein J is an integer and is more than or equal to 1 and less than or equal to 2K;
if the retrieval requirement only requires to carry out rough retrieval, J can take a value smaller than 2K, and if the retrieval requirement requires to carry out fine retrieval, J is 2K;
step S320, setting J equal to 0, J equal to or less than J;
in step S330, j equals j +1, the retrieval system combines the (j-1) -time filtered feature database with the jth component E of the vector EjPerforming feature matching, and filtering out the (j-1) filtered feature database and the component ejUnmatched feature data are obtained, and a feature database subjected to j times of filtering is obtained;
the feature database filtered for 0 times is the original feature database of the retrieval system;
step S340, judging whether J is smaller than J, if so, returning to the step S330, and if not, indicating that a final characteristic database meeting the retrieval requirement is obtained;
the values of the series J are different, the times of performing feature matching are different, that is, hierarchical retrieval can be performed by using hierarchical color distribution descriptors, for rough retrieval, only the first components of the vector E need to be extracted for feature matching (even only the first component of the vector E is extracted for feature matching), and only fine retrieval is performed for extracting all components of the vector E for feature matching, so that the requirement of efficiently accessing visual information is met;
only one feature (one component of the vector E) is used for each feature matching, and a large amount of feature data is filtered out for each feature matching, so that the calculated amount is greatly reduced, and the retrieval speed is improved;
step S400, the retrieval system searches out a corresponding image from an image database according to the final feature database and feeds back the searched image to the user;
and step S500, the user further searches according to the feedback result.
Claims (7)
1. An image content retrieval method based on a hierarchical color distribution descriptor, characterized by comprising the steps of:
step 1, a user inputs an image to be inquired and a retrieval requirement to a retrieval system;
step 2, the retrieval system constructs a hierarchical color distribution descriptor of the image to be queried;
step 2.1, the retrieval system divides the image to be inquired into K1×K2Each image block and calculating an average value of each image block, wherein K1And K2Are all even numbers;
when the image to be inquired is a gray image, the average value is the average value of all pixel gray values of the image block, and when the image to be inquired is a color image, the average value is the average value of all color component brightness values of the image block;
step 2.2, construct the directional mean vector C, C ═ C1，c2，c3，....，cK1×K2}；
If the average values of the image blocks of the image to be queried calculated in step 2.1 are not completely the same, step 2.2 specifically includes the following steps:
2.2.1, selecting a pair of maximum-minimum value pairs with the minimum spacing distance from the average values of the same circle of image blocks according to the sequence from outside to inside;
the step 2.2.1 is specifically as follows:
firstly, selecting a pair of average value maximum value-minimum value pairs with minimum spacing distance from the edge image blocks of the image to be inquired, and if the average value maximum value-minimum value pairs cannot be selected from the edge image blocks of the image to be inquired, selecting from the inner ring according to the sequence from outside to inside;
finding out all maximum values and minimum values from the average values of the same circle of image blocks, supposing that P maximum values and Q minimum values are found, forming P multiplied by Q maximum value-minimum value pairs, calculating the spacing distance S between each pair of maximum values and minimum values, and when calculating the spacing distance S, if the calculated value of the spacing distance S is larger than (2K)1+2K2-4)/2, so that S ═ 2K1+2K2-4)-S；
If only one maximum-minimum value pair with the minimum spacing distance exists through calculation, the maximum-minimum value pair is the maximum-minimum value pair with the minimum spacing distance selected;
if a plurality of maximum-minimum value pairs with the minimum spacing distance exist through calculation, a pair is determined by comparing adjacent items of the maximum value or the minimum value, and if the pair cannot be determined by comparing the adjacent items of the maximum value or the minimum value, the pair is selected randomly;
step 2.2.2, determining the rotation direction according to the maximum value-minimum value pair selected in the step 2.2.1;
the step 2.2.2 is specifically as follows:
if the maximum-minimum pair of values S selected in step 2.2.1 is less than (2K)1+2K24)/2, the direction of rotation is determined to be the direction from the maximum value to the minimum value over the minimum separation distance;
if the spacing distance S of the maximum value-minimum value pair selected in the step 2.2.1 is equal to (4K-4)/2, determining the rotation direction through adjacent items of the maximum value or the minimum value;
if the spacing distance S of the maximum value-minimum value pair selected in the step 2.2.1 is equal to (4K-4)/2 and the average values of other image blocks in the same circle are equal except the maximum value and the minimum value, selecting a pair of maximum value-minimum value pairs from the inner circle of the circle according to the sequence from outside to inside to determine the rotating direction;
step 2.2.3, traversing all image blocks of the image to be queried according to the spiral rotation method from outside to inside from the maximum value to the minimum value of the maximum value-minimum value pair selected in step 2.2.1 to the minimum value in the rotation direction determined in step 2.2.2, and forming the average values of the image blocks into a directional average value vector C according to the traversal sequence, wherein C is { C ═ C { (C) } C { (C {1，c2，c3，....，cK1×K2}；
Step 2.3, compress the directional mean vector and extract the front K of the compressed vector3One component constitutes a hierarchical color distribution descriptor E, E ═ E of the image to be queried1，e 2，e3，....，eK3In which K is3＜K1×K2；
Step 3, the retrieval system uses the hierarchical color distribution descriptor of the image to be queried to filter the feature database in a hierarchical manner until the retrieval requirement is met, and a final feature database is obtained;
step 4, the retrieval system searches out a corresponding image from the image database according to the final feature database and feeds back the searched image to the user;
and 5, the user further searches according to the feedback result.
2. The method for image content retrieval based on hierarchical color distribution descriptors of claim 1,
n1＝2，3，......；
n2＝2，3，......；K3＝K1+K2。
3. the method for retrieving image content based on hierarchical color distribution descriptor as claimed in claim 1, wherein if the average value of each image block of the image to be queried calculated in step 2.1 is equal, step 2.2 is implemented by making each component of the oriented average vector C equal to the average value.
4. The method for hierarchical color distribution descriptor-based image content retrieval according to claim 1, wherein the max-min value pair is replaced with a max-next-max value pair.
5. The method for hierarchical color distribution descriptor-based image content retrieval according to claim 1, wherein the max-min value pair is replaced with a min-next-min value pair.
6. The method for hierarchical color distribution descriptor-based image content retrieval according to claim 1, wherein said oriented mean vector C is compressed in said step 2.3 using HAAR wavelet compression.
7. The method for retrieving image content based on hierarchical color distribution descriptors as claimed in claim 1, wherein said step 3 comprises the following steps:
step 3.1, the retrieval system determines the stage number J of the hierarchical retrieval according to the retrieval requirement input in the step 1, wherein J is an integer and J is more than or equal to 1 and less than or equal to K3；
Step 3.2, setting J to be 0, and setting J to be less than or equal to J;
step 3.3, j equals j +1, the retrieval system combines the (j-1) filtered feature database with the jth component E of the vector EjPerforming feature matching, and filtering out the (j-1) filtered feature database and the component ejUnmatched feature data are obtained, and a feature database subjected to j times of filtering is obtained;
and 3.4, judging whether J is smaller than J, if so, returning to the step 3.3, and if not, indicating that the final characteristic database meeting the retrieval requirement is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110249085 CN102306179B (en) | 2011-08-26 | 2011-08-26 | Image content retrieval method based on hierarchical color distribution descriptor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110249085 CN102306179B (en) | 2011-08-26 | 2011-08-26 | Image content retrieval method based on hierarchical color distribution descriptor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102306179A CN102306179A (en) | 2012-01-04 |
| CN102306179B true CN102306179B (en) | 2012-12-05 |
Family
ID=45380041
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110249085 Active CN102306179B (en) | 2011-08-26 | 2011-08-26 | Image content retrieval method based on hierarchical color distribution descriptor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102306179B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10169803B2 (en) | 2014-06-26 | 2019-01-01 | Amazon Technologies, Inc. | Color based social networking recommendations |
| US9996579B2 (en) * | 2014-06-26 | 2018-06-12 | Amazon Technologies, Inc. | Fast color searching |
| US9785649B1 (en) | 2014-09-02 | 2017-10-10 | Amazon Technologies, Inc. | Hue-based color naming for an image |
| CN111680183B (en) * | 2020-08-13 | 2020-11-24 | 成都睿沿科技有限公司 | Object retrieval method and device, storage medium and electronic equipment |
| CN117115481B (en) * | 2023-04-14 | 2024-09-27 | 荣耀终端有限公司 | Positioning method, electronic device and medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101526944A (en) * | 2008-12-23 | 2009-09-09 | 广州乐庚信息科技有限公司 | Image retrieving comparison method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4208780B2 (en) * | 2004-07-07 | 2009-01-14 | キヤノン株式会社 | Image processing system, control method for image processing apparatus, and program |
-
2011
- 2011-08-26 CN CN 201110249085 patent/CN102306179B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101526944A (en) * | 2008-12-23 | 2009-09-09 | 广州乐庚信息科技有限公司 | Image retrieving comparison method |
Non-Patent Citations (2)
| Title |
|---|
| 刘金梅.基于分级颜色分布描述符的图像内容检索方法.《中国优秀硕士论文电子期刊网》.2004,摘要、正文第14页倒数第2段. |
| 基于分级颜色分布描述符的图像内容检索方法;刘金梅;《中国优秀硕士论文电子期刊网》;20041102;摘要、正文第14页倒数第2段 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102306179A (en) | 2012-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kayhan et al. | Content based image retrieval based on weighted fusion of texture and color features derived from modified local binary patterns and local neighborhood difference patterns | |
| CN103544216B (en) | The information recommendation method and system of a kind of combination picture material and keyword | |
| Ahmadi et al. | Context-aware saliency detection for image retargeting using convolutional neural networks | |
| CN101477529B (en) | Three-dimensional object retrieval method and apparatus | |
| CN102306179B (en) | Image content retrieval method based on hierarchical color distribution descriptor | |
| CN103761295A (en) | Automatic picture classification based customized feature extraction algorithm for art pictures | |
| Kavitha et al. | An efficient content based image retrieval using color and texture of image sub blocks | |
| Hussain et al. | Robust pre-processing technique based on saliency detection for content based image retrieval systems | |
| Atlam et al. | Comparative study on CBIR based on color feature | |
| Chen et al. | HyNet: A novel hybrid deep learning approach for efficient interior design texture retrieval | |
| CN106203461A (en) | An image processing method and device | |
| CN103995864B (en) | A kind of image search method and device | |
| Xu et al. | Relevance feedback for Content-based Image Retrieval using deep learning | |
| CN108319959A (en) | A kind of corps diseases image-recognizing method compressed based on characteristics of image with retrieval | |
| Latha et al. | Enhanced hybrid CBIR based on multichannel LBP oriented color descriptor and HSV color statistical feature | |
| CN114818903A (en) | Image classification method and system | |
| Yakin et al. | Application of content based image retrieval in digital image search system | |
| Manjusha et al. | Content based image retrieval using wavelet transform and feedback algorithm | |
| US20120131026A1 (en) | Visual information retrieval system | |
| Latha et al. | Hybrid CBIR method using statistical, DWT‐Entropy and POPMV‐based feature sets | |
| CN106570125B (en) | Remote sensing image retrieval method and device for rotational scaling and translation invariance | |
| Singh et al. | An effective mechanism to neutralize the semantic gap in content based image retrieval (CBIR). | |
| Sreedevi et al. | Fast image retrieval with feature levels | |
| Saeed et al. | Content-based image retrieval by multi-features extraction and k-means clustering | |
| Lee et al. | Content-based image retrieval using spatial-color and Gabor texture on a mobile device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| DD01 | Delivery of document by public notice |
Addressee: Wu guanyong Document name: payment instructions |
|
| DD01 | Delivery of document by public notice | ||
| DD01 | Delivery of document by public notice |
Addressee: Wu Guanyong Document name: Notice of patent termination |
|
| DD01 | Delivery of document by public notice |