Intelligent image similarity comparison method fusing channel and texture features

Abstract

The invention provides an intelligent image similarity comparison method fusing channel and texture features. The method comprises the following steps: respectively carrying out the following operations on a first image and a second image: carrying out first processing on the images to obtain channel pixel features of the images; performing second processing on the image to obtain texture features of the image; performing third processing on the channel pixel feature and the texture feature to obtain a first vector; performing fourth processing on the first vector to obtain a second vector; and performing calculation according to the second vector of the first image and the second vector of the second image to obtain the similarity degree of the first image and the second image. According to the intelligent image similarity comparison method fusing the channel and the texture features, two feature indexes related to image similarity comparison, namely the texture features of the image and the channel pixel features, are fused; the problems of low estimation accuracy, algorithm compression distortion and low performance of current mainstream image pairing query and similarity quantitative detection can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image similarity comparison, and in particular to an image similarity intelligent comparison method integrating channel and texture features. Background Art

[0002] In the existing image pairing query and similarity quantification detection tasks, the mainstream detection schemes include histogram pixel number distribution description method, hash spectrum decomposition approximation method, and structural similarity comparison method. The histogram pixel number distribution description method is simple and easy to use and has a small amount of calculation. This method counts the number of histogram pixels in segments based on the similar histogram grayscale pixel interval distribution of similar images and the grayscale pixel interval statistics of the entire image; the hash spectrum decomposition approximation method has the characteristics of removing high-frequency noise and performing discrete low-frequency filtering. This method selects images of any size as input and performs normalization and discrete cosine transform processing, so as to perform similarity comparison analysis in the low-frequency band; the structural similarity comparison method has the characteristics of high fitting accuracy, which comprehensively determines the pixel statistical characteristics of the image through the mean and variance of the grayscale pixels of the given image.

[0003] However, the histogram pixel distribution description method calculates the absolute value distribution of image pixels, but does not contain the absolute or relative position information of image elements. The low similarity recall rate makes it difficult to be widely used; the hash spectrum decomposition approximation method only focuses on the low-frequency information in the image and ignores the high-frequency information. The high-frequency component contains the detailed texture information of the image, which makes it easy to cause information distortion in the recognition process and difficult to meet the requirements in engineering practice; the structural similarity comparison method uses the mean and variance of the grayscale level to approximate the high-frequency texture information of the image, but its high calculation delay also limits its use. Summary of the invention

[0004] The object of the present invention is to solve at least one of the technical drawbacks.

[0005] To this end, an object of the present invention is to propose an intelligent image similarity comparison method that integrates channel and texture features to solve the problems mentioned in the background technology and overcome the deficiencies in the prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] An intelligent image similarity comparison method integrating channel and texture features comprises: performing the following operations on a first image and a second image respectively:

[0008] Performing a first processing on the image to obtain channel pixel features of the image;

[0009] Performing a second processing on the image to obtain texture features of the image;

[0010] Performing a third process on the channel pixel feature and the texture feature to obtain a first vector;

[0011] Performing a fourth process on the first vector to obtain a second vector;

[0012] The similarity between the first image and the second image is obtained by calculating according to the second vector of the first image and the second vector of the second image.

[0013] Preferably, the first processing of the image to obtain the channel pixel features of the image includes: performing image channel Channel pixel filtering decomposition and transformation on the image to obtain the channel pixel features of the image, specifically, performing dimensionality reduction and pixel filtering decomposition and transformation on the three-channel pixel values ​​of each pixel of the image by means of low-pass mean filtering to obtain the channel pixel features of the image.

[0014] Preferably, the second processing of the image to obtain the texture features of the image includes: performing texture feature convolution kernel detection and binarization processing on the image to obtain the texture features of the image. Specifically, edge features of the image are extracted using the Canny edge operator, and a boundary detection threshold is set, and the edge feature matrix value set {0, 255} is mapped to {0, 1} to obtain the texture features of the image.

[0015] Preferably, the third processing of the channel pixel features and the texture features to obtain the first vector includes: performing feature vector fusion and dimensionality reduction compression on the channel pixel features and the texture features of the image to obtain a feature vector of optimal length. Specifically, the image values ​​of the pixels corresponding to the channel pixel features and the texture features of the image are added to obtain a fused feature, and the fused feature is reshaped and compressed by sliding convolution to obtain a feature vector of optimal length.

[0016] Preferably, performing the fourth processing on the first vector to obtain the second vector comprises: performing a normalization operation on the optimal length feature vector to obtain a reduced feature vector.

[0017] As a preference, the detection image resolution length is w, the height is h, and the optimal compression length is l * , the vector representation after fusion feature compression The global vector mean is μ, and the global vector standard deviation is σ, then the sorting i and normalization operations are as follows:

[0018]

[0019] i∈{0, 1, 2, ..., M}

[0020] v * =(v i -μ) / σ

[0021] Among them, v * Represents the value of the normalized reduced eigenvector.

[0022] Preferably, the similarity between the first image and the second image obtained by calculating based on the second vector of the first image and the second vector of the second image includes: calculating the vector angle using the cosine similarity method based on the second vector of the first image and the second vector of the second image, wherein the smaller the angle is, the higher the similarity between the first image and the second image is.

[0023] Preferably, the reduced feature vector of the first image is set to be The reduced eigenvector of the second image is V j * ={[v 1 , v 2 ,...v M ]}, the vector angle calculation formula is:

[0024]

[0025] Preferably, the first image is an image to be inspected, and the second image is an image in a standard metadata image library.

[0026] Therefore, the present invention has the following beneficial effects:

[0027] The present invention discloses an intelligent image similarity comparison method that integrates channel and texture features, which integrates two feature indicators related to image similarity comparison, namely, image texture features and channel pixel features. First, in the image comparison process, the three-channel pixel data of the image to be tested is collected in real time, and the channel pixel features are extracted by dimensionality reduction using the position-calibrated three-channel mean filtering method; secondly, the extracted binary edge texture features are reduced in dimension via one-dimensional sliding convolution and fused element-by-element with the channel pixel features extracted by dimensionality reduction to obtain the final feature vector representation of the image to be tested; finally, in order to improve the comparison efficiency, the standard metadata image library images to be compared are pre-extracted and pre-compressed, and in subsequent comparisons, they are directly queried through database serialization to improve the real-time retrieval speed. The method of the present invention can effectively solve the problems of low estimation accuracy, algorithm compression distortion and low performance of current mainstream image pairing queries and similarity quantization detection.

[0028] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0030] Figure 1 It is the overall flow chart of the method of the present invention;

[0031] Figure 2 2 is a schematic diagram of pixel filtering decomposition and transformation of an image channel according to an embodiment of the present invention;

[0032] Figure 3 Schematic diagram of texture feature convolution kernel detection and binarization according to an embodiment of the present invention;

[0033] Figure 4 is a schematic diagram of a process of feature vector compression according to an embodiment of the present invention;

[0034] Figure 5 is a schematic diagram of optimal compression ratio quantization coefficients according to an embodiment of the present invention;

[0035] Figure 6 is a schematic diagram of algorithm recall rate curves of different numbers of test sets according to an embodiment of the present invention;

[0036] Figure 7 It is a schematic diagram of the structure of the system of the embodiment of the present invention. DETAILED DESCRIPTION

[0037] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0038] like Figure 1 As shown, a method for intelligently comparing image similarity by integrating channel and texture features includes: performing the following operations on a first image and a second image respectively:

[0039] Performing a first processing on the image to obtain channel pixel features of the image;

[0040] Performing a second processing on the image to obtain texture features of the image;

[0041] Performing a third process on the channel pixel feature and the texture feature to obtain a first vector;

[0042] Performing a fourth process on the first vector to obtain a second vector;

[0043] The similarity between the first image and the second image is obtained by calculating according to the second vector of the first image and the second vector of the second image.

[0044] The present invention provides an intelligent image similarity comparison method integrating channel and texture features, which integrates two feature indicators related to image similarity comparison, namely, image texture features and channel pixel features. It can effectively solve the problems of low estimation accuracy, algorithm compression distortion and low performance in current mainstream image pairing query and similarity quantization detection.

[0045] Preferably, performing a first processing on the image to obtain the channel pixel features of the image includes: performing image channel Channel pixel filtering decomposition and transformation on the image to obtain the channel pixel features of the image, specifically, performing dimensionality reduction and pixel filtering decomposition and transformation on the three-channel pixel values ​​of each pixel of the image by means of low-pass mean filtering to obtain the channel pixel features of the image.

[0046] The present invention performs dimensionality reduction and pixel filtering decomposition transformation on the three-channel pixel values ​​of each pixel of the image by low-pass mean filtering, thereby reducing the computational complexity while maintaining the image channel feature information.

[0047] Preferably, performing a second processing on the image to obtain the texture features of the image includes: performing texture feature convolution kernel detection and binarization processing on the image to obtain the texture features of the image. Specifically, using the Canny edge operator to extract edge features of the image, setting a boundary detection threshold, and mapping the edge feature matrix value set {0, 255} to {0, 1} to obtain the texture features of the image.

[0048] Texture features are the most direct features in image similarity comparison. Texture features refer to pixels whose pixel differences are not 0, which represent the content structure information of the image. Mapping the edge feature matrix value set {0, 255} to {0, 1} to realize the binarization of the edge feature matrix can prevent the gradient explosion phenomenon.

[0049] Preferably, performing a third processing on the channel pixel features and the texture features to obtain the first vector includes: performing feature vector fusion and dimensionality reduction compression on the channel pixel features and the texture features of the image to obtain a feature vector of optimal length. Specifically, performing addition calculation on the image values ​​of the pixels corresponding to the channel pixel features and the texture features of the image to obtain a fused feature, and performing reshaping and sliding convolution compression on the fused feature to obtain a feature vector of optimal length.

[0050] The length of the feature dimension is a proportional function of the image resolution, and the resolution of the image is generally in the millions or tens of millions, which can easily cause the disaster of dimensionality. Therefore, it is necessary to perform dimensionality reduction and compression processing on the channel pixel features and texture feature vectors respectively.

[0051] Preferably, performing the fourth processing on the first vector to obtain the second vector includes: performing a normalization operation on the optimal length feature vector to obtain a reduced feature vector.

[0052] The efficiency of image feature similarity comparison is improved by normalizing the feature vector with the optimal length.

[0053] It can be understood that the first vector is the optimal length feature vector, and the second vector is the reduced feature vector.

[0054] As a preferred method, let the detection image resolution length be w, the height be h, and the optimal compression length be l * , the vector representation after fusion feature compression The global vector mean is μ, and the global vector standard deviation is σ, then the sorting i and normalization operations are as follows:

[0055]

[0056] i∈{0, 1, 2, ..., M}

[0057] v * =(v i -μ) / σ

[0058] Among them, v * Represents the value of the normalized reduced eigenvector.

[0059] Preferably, calculating the degree of similarity between the first image and the second image based on the second vector of the first image and the second vector of the second image includes: calculating the vector angle using the cosine similarity method based on the second vector of the first image and the second vector of the second image, wherein the smaller the angle, the higher the similarity between the first image and the second image.

[0060] As a preference, let the reduced feature vector of the first image be The reduced eigenvector of the second image is V j * ={[v 1 ,v 2 ,...v M ]}, the vector angle calculation formula is:

[0061]

[0062] Preferably, the first image is an image to be inspected, specifically an iptv streaming video frame, and the second image is an image in a standard metadata image library.

[0063] In the following embodiments, an image similarity intelligent comparison method integrating channels and texture features of the present invention is described in detail with reference to the accompanying drawings.

[0064] In the existing image pairing query and similarity quantification detection tasks, the mainstream detection schemes include histogram pixel number distribution description method, hash spectrum decomposition approximation method, and structural similarity comparison method.

[0065] The histogram pixel distribution description method calculates the absolute value distribution of image pixels but does not contain the absolute or relative position information of image elements. The low similarity recall rate makes it difficult to be widely used; the hash spectrum decomposition approximation method only focuses on the low-frequency information in the image and ignores the high-frequency information, while the high-frequency component contains the detailed texture information of the image, which makes it easy to cause information distortion in the recognition process and difficult to meet the requirements in engineering practice; the structural similarity comparison method uses the mean and variance of the gray level to approximate the high-frequency texture information of the image, but its high calculation delay also limits its use. Therefore, the present invention proposes an image similarity comparison method that integrates channel and texture features.

[0066] Generally speaking, from the perspective of image intuitive morphology, similar images tend to be consistent in texture structure, brightness and contrast; and from the perspective of technical indicators, similar images should also have certain robustness constraints under small affine transformations, that is, it is completely feasible to integrate the vector compression and dimensionality reduction of image texture structure and channel features and compare them in the same kernel space.

[0067] The method of the present invention integrates two feature indicators related to image similarity comparison, namely, image texture edge features and channel pixel features. First, in the image comparison process, the three-channel pixel data of the image to be tested is collected in real time, and the channel pixel features are extracted by dimensionality reduction using the position-calibrated three-channel mean filtering method; secondly, the extracted binary edge texture features are reduced in dimension through one-dimensional sliding convolution and fused element by element with the channel pixel features extracted by dimensionality reduction to obtain the final feature vector representation of the image to be tested; finally, in order to improve the comparison efficiency, the standard metadata image library images to be compared are pre-extracted and pre-compressed, and in the subsequent comparison, the database serialization query is directly used to improve the real-time retrieval speed.

[0068] The method of the present invention can effectively solve the problems of low estimation accuracy, algorithm compression distortion and low performance of current mainstream management methods, and is an innovation and effective supplement to traditional monitoring methods.

[0069] The present invention is used to realize the image similarity comparison task based on the existing standard metadata image library, and mainly uses the image channel pixel filtering decomposition and transformation, texture feature convolution kernel detection and binarization, feature vector fusion dimensionality reduction compression, layer normalization and cosine angle similarity technology. The method of the present invention first uses FFmpeg technology to intercept the target video from the standard metadata image library, and then performs three-channel channel filtering, edge feature extraction, and feature compression and fusion operations on the metadata image in the library. Finally, the IPTV streaming video frame image to be inspected is compared with the metadata image in the library after channel filtering, texture fusion and normalization, and the library image with the greatest similarity to the IPTV streaming video frame is retrieved, and it is used to determine whether the IPTV streaming video frame image to be inspected exists in the metadata image library. The specific steps include:

[0070] (1) Identify the monitored objects and their indicators:

[0071] Monitored object: Here it refers to the IPTV streaming video frame compared with the standard metadata image library;

[0072] Estimation indicators: The similarity of structural distribution and pixel distribution between IPTV streaming video frames within the statistical range and the standard metadata image library.

[0073] (2) Image channel pixel filtering decomposition and transformation:

[0074] The image values ​​of the three channels of an image all carry characteristic information, but if the image values ​​of all channels of all pixels of the image are calculated, there will be the problem of dimensionality curse, and the space-time complexity of the calculation will increase exponentially. In order to solve this problem, the present invention adopts a method similar to the image grayscale value calculation, and performs dimensionality reduction and pixel filtering decomposition and transformation on the three-channel pixel values ​​of each pixel of the image through low-pass mean filtering, while maintaining the characteristic information of the image channel, reducing the calculation complexity. The operation diagram is shown in the figure. Figure 2 shown.

[0075] (3) Texture feature convolution kernel detection and binarization:

[0076] Texture feature is the most direct feature in image similarity comparison. Texture feature refers to the pixel point whose pixel difference is not 0, which represents the content structure information of the image. The present invention uses Canny edge operator to extract edge features of the image and set a reasonable boundary detection threshold. Furthermore, in order to facilitate subsequent normalization and fusion operations, the edge feature matrix value set {0, 255} is mapped to {0, 1}. The flow chart is as follows: Figure 3 shown.

[0077] The edge feature extraction operator uses the Canny function of the open source visual processing library Opencv. Images with different resolutions and visual qualities have different edge feature matrices. To prevent the gradient explosion phenomenon, {0, 255} is mapped to {0, 1} to achieve the binarization of the edge feature matrix. The Canny function edge feature extraction belongs to the category of open source algorithms, and its specific algorithm does not belong to the innovative point of the present invention and the scope of application protection, so the present invention will not be repeated here.

[0078] (4) Feature vector fusion dimensionality reduction compression:

[0079] After step (1), step (2) and step (3), the channel pixel features and texture features of the standard metadata image library and the IPTV streaming video frame are obtained through the image channel pixel filtering decomposition transformation, texture feature convolution kernel detection and binarization technology respectively. The feature dimension length is a positive proportional function of the image resolution and the resolution of the image is generally in the millions or tens of millions, which is very easy to cause dimensionality disaster. The two groups of feature vectors need to be compressed by dimensionality reduction respectively. The optimal ratio coefficient of dimensionality reduction compression is obtained by binary division. The main process of compression of the two groups of feature vectors and the schematic diagram of the optimal compression ratio quantization coefficient are shown as follows. Figure 4 and Figure 5 shown.

[0080] (5) Layer Normalization:

[0081] In order to improve the efficiency of image feature similarity comparison, the optimal length feature vector needs to be normalized, that is, one-dimensional convolution and feature vector reduction are performed. Here, the mean variance reduction method is adopted. Assume that the detection image resolution length is w, the height is h, and the optimal compression length is l * , the vector representation after fusion feature compression The global vector mean is μ, and the global vector standard deviation is σ. Then the sorting i and normalization operations are shown in the following formulas (1), (2), and (3), respectively:

[0082]

[0083] i∈{0,1,2,...,M} (2)

[0084] v * =(v i -μ) / σ (3)

[0085] Where M is determined by whether it can be divided by the image resolution. * Represents the value of the normalized reduced eigenvector.

[0086] (6) Cosine similarity calculation

[0087] After completing step (5), that is, the standard metadata image library and the IPTV streaming video frame image are converted into the specification vector of the same feature space, in order to determine the relationship between the IPTV streaming video frame image and the image in the standard metadata image library, the cosine similarity method is used to detect the vector angle. The central idea of ​​this method is that for the quantitative representation of similar image feature vectors, their angle can be determined by the inverse cosine trigonometric function of the ratio of their vector dot product and the square root product of their respective vector norms. The closer the angle is to 0, the more similar the two image feature vectors are, and vice versa. If the feature vector of the image to be tested is The image feature vector of the standard metadata image library is V j * ={[v 1 , v 2 ,...v M ]}The angle calculation formula is shown in formula (4):

[0088]

[0089] Select the test set image list [100, 200, 300, 400, 500, 600, 700, 800, 900, 1000] and 7000 images of the standard metadata image library to test the recall rate of the algorithm. The recall rate curve is as follows Figure 6 shown.

[0090] Depend on Figure 6 It can be seen that as the number of test set images increases, the recall rate of the algorithm also changes accordingly, but the range of the change is relatively stable. The maximum value of the interval is 0.988, and the minimum value is 0.953. The minimum value occurs when the number of test set images is 300, and the maximum value occurs when the number of test set images is 1000, which shows the dynamic effectiveness and accuracy of the algorithm of the present invention.

[0091] The present invention discloses an intelligent image similarity comparison method that integrates channel and texture features, which integrates two feature indicators related to image similarity comparison, namely, image texture features and channel pixel features. First, in the image comparison process, the three-channel pixel data of the image to be tested is collected in real time, and the channel pixel features are extracted by dimensionality reduction using the position-calibrated three-channel mean filtering method; secondly, the extracted binary edge texture features are reduced in dimension via one-dimensional sliding convolution and fused element-by-element with the channel pixel features extracted by dimensionality reduction to obtain the final feature vector representation of the image to be tested; finally, in order to improve the comparison efficiency, the standard metadata image library images to be compared are pre-extracted and pre-compressed, and in subsequent comparisons, they are directly queried through database serialization to improve the real-time retrieval speed. The method of the present invention can effectively solve the problems of low estimation accuracy, algorithm compression distortion and low performance of current mainstream image pairing queries and similarity quantization detection.

[0092] This embodiment also provides an image similarity intelligent comparison system that integrates channel and texture features, such as Figure 7 As shown, including:

[0093] Image acquisition module: used to obtain the first image and the second image data, namely the standard metadata image library image and the IPTV streaming video frame image.

[0094] The first processing module is used to perform image channel pixel filtering decomposition and transformation on the image to obtain the channel pixel features of the image. Specifically, the three-channel pixel values ​​of each pixel of the image are dimensionally reduced and pixel filtered and decomposed and transformed by low-pass mean filtering to obtain the channel pixel features of the image.

[0095] The second processing module is used to perform texture feature convolution kernel detection and binarization processing on the image to obtain the texture features of the image. Specifically, the Canny edge operator is used to extract the edge features of the image, and the boundary detection threshold is set to map the edge feature matrix value set {0, 255} to {0, 1} to obtain the texture features of the image.

[0096] The third processing module is used to perform feature vector fusion and dimensionality reduction compression on the channel pixel features and texture features of the image to obtain a feature vector of optimal length. Specifically, the image values ​​of the pixels corresponding to the channel pixel features and texture features of the image are added to obtain the fused features, and the fused features are reshaped and compressed by sliding convolution to obtain the feature vector of optimal length.

[0097] The fourth processing module is used to perform a normalization operation on the optimal length feature vector to obtain a reduced feature vector.

[0098] Angle calculation module: used to calculate the vector angle using the cosine similarity method according to the second vector of the first image and the second vector of the second image.

[0099] This embodiment also provides a computer device, which is applicable to the above-mentioned method for intelligently comparing image similarity by fusing channels and texture features, and includes: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the steps of the method for intelligently comparing image similarity by fusing channels and texture features as proposed in the above-mentioned embodiment.

[0100] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0101] The present embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, an intelligent image similarity comparison method for fusing channels and texture features as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0102] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0103] It is not difficult for those skilled in the art to understand that the present invention includes any combination of the invention content and specific implementation methods of the above specification and the various parts shown in the drawings. Due to the limited space and to make the specification concise, the various schemes composed of these combinations are not described one by one. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0104] Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. Those skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and purpose of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent image similarity comparison method integrating channel and texture features, characterized in that: include: Perform the following operations on the first image and the second image respectively: Performing a first processing on the image to obtain channel pixel features of the image; Performing a second processing on the image to obtain texture features of the image; Performing a third process on the channel pixel feature and the texture feature to obtain a first vector; Performing a fourth process on the first vector to obtain a second vector; The similarity between the first image and the second image is obtained by calculating according to the second vector of the first image and the second vector of the second image.

2. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 1, characterized in that: The first processing of the image to obtain the channel pixel features of the image includes: performing image channel Channel pixel filtering decomposition and transformation on the image to obtain the channel pixel features of the image. Specifically, the three-channel pixel values ​​of each pixel of the image are dimensionally reduced and pixel filtered and decomposed and transformed by low-pass mean filtering to obtain the channel pixel features of the image.

3. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 1, characterized in that: The second processing of the image to obtain the texture features of the image includes: performing texture feature convolution kernel detection and binarization processing on the image to obtain the texture features of the image. Specifically, using the Canny edge operator to extract edge features of the image, setting a boundary detection threshold, and mapping the edge feature matrix value set {0, 255} to {0, 1} to obtain the texture features of the image.

4. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 1, characterized in that: The third processing of the channel pixel features and the texture features to obtain the first vector includes: performing feature vector fusion and dimensionality reduction compression on the channel pixel features and the texture features of the image to obtain a feature vector of optimal length. Specifically, the image values ​​of the pixels corresponding to the channel pixel features and the texture features of the image are added to obtain the fused features, and the fused features are reshaped and compressed by sliding convolution to obtain the feature vector of optimal length.

5. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 1, characterized in that: The performing the fourth processing on the first vector to obtain the second vector includes: performing a normalization operation on the optimal length feature vector to obtain a reduced feature vector.

6. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 5, characterized in that: Assume the detection image resolution length is w, the height is h, and the optimal compression length is l * , the vector representation V after fusion feature compression i ={[v1,v2,...v l* ]}, the global vector mean is μ, and the global vector standard deviation is σ, then the sorting i and normalization operations are as follows: i∈{0,1,2,…,M} v * =(v i -m) / s Among them, v * Represents the value of the normalized reduced eigenvector.

7. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 1, characterized in that: The method of calculating the similarity between the first image and the second image based on the second vector of the first image and the second vector of the second image includes: calculating the vector angle using the cosine similarity method based on the second vector of the first image and the second vector of the second image, and the smaller the angle, the higher the similarity between the first image and the second image.

8. The method for intelligently comparing image similarity by integrating channel and texture features according to claim 7, characterized in that: Suppose the reduced eigenvector of the first image is The reduced eigenvector of the second image is The vector angle calculation formula is:

9. The method for intelligently comparing image similarity by integrating channel and texture features according to any one of claims 1 to 8, characterized in that: The first image is an image to be inspected, and the second image is an image in a standard metadata image library.

Images