Image retrieval system based on similar images
By using image segmentation and feature area extraction technologies in the image retrieval system, the problem of inefficient image retrieval is solved, and a more balanced search rate and higher accuracy are achieved.
Patent Information
- Application Number
- CN202410958710.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Due to the influence of information islands and search methods, the image retrieval efficiency is low and the classification logic and visual characteristics of clothing images cannot be optimized, resulting in unsatisfactory search results, and the image database is large in size and complex in resolution, which consumes the search process. Duration and uneven rate.
Using an image retrieval system based on similar images, the image data is distinguished, stored and featured image block extraction by combining control terminals, storage layers, index layers and search layers, and using image segmentation and featured area extraction technology to improve search efficiency and accuracy sex.
The rate of similar image retrieval process is achieved more balanced and stable, which improves the search efficiency and accuracy, and ensures that the characteristic position similarity between similar images and the search target is higher.
Smart Images

Figure CN118916505A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image retrieval, and in particular to an image retrieval system based on similar images. Background Art
[0002] Similar image retrieval, also known as "image search" or "reverse image search engine", is an important research direction in the field of computer vision. It refers to finding other similar images in a large number of image databases based on a given image.
[0003] The invention patent with application number 201911301577.5 discloses a similar image retrieval system based on a deep convolutional neural network. The image retrieval system is connected to the Internet and is characterized in that it includes: a sample generation module; a comparison module; a training module; and an auxiliary screening module; wherein the sample generation module is used to identify and analyze the image feature information of random samples, by calculating the sample image feature vector and forming the sample image feature vector library, the comparison module compares the image feature vectors to obtain similar image information, wherein the training module uses the LiftedStruct loss function to train sample data to form a first algorithm model, and the auxiliary screening module is used to label similar images.
[0004] The application aims to solve the problem that "with the development of network communications, image content-based retrieval is becoming more and more popular. However, due to the influence of information islands and retrieval methods, the existing image retrieval is inefficient and the results cannot meet the needs. For example, in the retrieval needs for clothing, the classification logic and visual features unique to clothing images cannot be optimized, resulting in retrieval results that cannot meet user expectations. In the field of image retrieval, deep learning technology usually uses loss functions to train training sample sets. Commonly used loss functions include: Contrastive loss function, Triplet loss function, Lifted Structloss function, etc. In traditional image retrieval schemes, global features of images such as color, texture, shape or local features such as SIFT, SURF or a combination of the two are often used as image vectors to measure the distance between images. In the actual retrieval process, the output images may be very similar or identical. In other words, traditional deep learning-based image retrieval only inputs the processed image data into the deep network for training, has weak background generalization ability, and usually has insufficient training data, and the output data accuracy is not high."
[0005] However, in the current process of using images to search image databases, due to the huge volume of images stored in image databases, complex image resolution, and diverse characteristics, the image search process is often time-consuming and the overall search rate balance is poor.
[0006] To this end, we propose an image retrieval system based on similar images. Summary of the invention
[0007] In view of the above-mentioned shortcomings of the prior art, the present invention provides an image retrieval system based on similar images, which solves the technical problems raised in the above-mentioned background technology.
[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:
[0009] An image retrieval system based on similar images, comprising: a control terminal, a storage layer, an index layer and a retrieval layer;
[0010] The control terminal is the main control terminal of the system and is used to issue execution commands;
[0011] The image data to be stored are uploaded through the storage layer and stored separately. The image data in each differentiated storage interval is differentiated again in the differentiated storage interval based on the image resolution. The index layer traverses the image data stored in each differentiated storage interval in the storage layer, extracts index image blocks on each group of image data, and marks the source path for each extracted index image block. The retrieval layer uploads a group of image data as a similar image retrieval target, extracts feature image blocks based on the index image block extraction logic in the similar image retrieval target, searches for similar index image blocks in the index image blocks extracted in the similar image retrieval target through the feature image blocks extracted, and simultaneously obtains the source path of the found index image block, determines the image data corresponding to the index image block based on the source path, and outputs the determined image data as the similar image data of the similar image retrieval target;
[0012] The index layer includes a receiving module, an extracting module and a marking module, wherein the receiving module is used to continuously receive the image data stored in the storage layer, the extracting module is used to synchronously obtain the image data received by the receiving module, extract the index image blocks from the image data, and the marking module is used to continuously receive the index image blocks extracted by the extracting module, and mark the source path of the index image blocks;
[0013] When the extraction module extracts the index image block from the image data, it first performs segmentation processing on the image data, and further performs the index image block extraction operation using the sub-images obtained by segmentation;
[0014] The image data segmentation process is subject to:
[0015]
[0016] Where: λ is the complexity of the image data; M and N are the number of rows and columns of the image data; δ x f(i,j) is the first-order difference of pixel (i,j) in the x direction; δ y f(i,j) is the first-order difference of pixel (i,j) in the y direction; E h is the horizontal symmetric error of the image data; E v is the vertical symmetric error of the image data; g is the image data segmentation parameter; γ is the adjustment factor;
[0017] The greater the complexity of the image data, the more information the image data contains. Conversely, the less information the image data contains. The image data segmentation parameter g is rounded up using the rounding method. After the image data segmentation parameter g is obtained, the number of image data segments is g. 2 , and the size of each segmented sub-image is equal.
[0018] Furthermore, the storage layer includes a storage module, a distinguishing module, and a grouping module. The storage module is used to upload image data and store the uploaded image data. The distinguishing module is used to traverse the image data stored in the storage module, read the image data names, and based on the total amount of different image data names, construct a distinguishing storage interval equal to the total amount of image names in the storage module, and store image data with the same image name in the same distinguishing storage interval. The grouping module is used to identify the resolution of each image data stored in each distinguishing storage interval in the storage module, and perform a distinguishing operation on each image data stored in the distinguishing storage interval again according to the image data resolution.
[0019] Each set of image data uploaded in the storage module is labeled with a name, including: people, animals, scenery, objects, natural scenery, and weather.
[0020] Furthermore, during the operation phase of the grouping module, after identifying the resolution of each image data in each differentiated storage interval, the image data stored in each differentiated storage interval is arranged in descending or ascending order based on the resolution of the image data, and the first and last groups of image data in the image data arranged continuously in each differentiated storage interval are selected as a set of subsets;
[0021] When the number of image data stored in the distinguished storage interval is odd, a middle group of image data in the distinguished storage interval arranged in descending or ascending order based on the image data resolution is taken as a subset.
[0022] Furthermore, in the operation phase of the receiving module, each subset in each storage interval of the storage module is taken as a receiving target, and the receiving module continuously receives the image data stored in the storage layer, that is, continuously receives the subset in the storage module;
[0023] When the marking module marks the source path of the index image block, the subset where the source image data of the index image block is located and the location of the differentiated storage interval where the subset is located are used as marking content to perform the marking operation of the source path of the index image block.
[0024] Furthermore, the horizontal symmetric error of the image data and the vertical symmetric error of the image data are obtained by the following formula:
[0025]
[0026] In the formula: |I(x,y)-I h (x, y) is the original image I and the horizontal symmetric image I h The absolute value of the difference in pixel values at (x, y); |I(x, y)-I v (x,y) is the original image I and the vertical symmetric image I v The absolute value of the difference in pixel values at (x, y);
[0027] Among them, the horizontally symmetric image I h That is, the image obtained by flipping the original image along the horizontal central axis, the vertically symmetrical image I v That is, the image obtained by flipping the original image along the vertical central axis.
[0028] Furthermore, the adjustment factor γ≥1, and is subject to the setting logic that the higher the image data resolution, the larger the adjustment factor, and vice versa;
[0029]
[0030] Where: f(i,j) is the pixel value of the image data at the coordinate position (i,j).
[0031] Furthermore, in the stage of extracting the index image block, the extraction module takes all sub-images obtained by segmenting the image data as a set, and extracts sub-images from the set as index image blocks:
[0032]
[0033] Where: DIFF a is the comprehensive difference between sub-image a and other sub-images in the set; Q is the total number of sub-images in the set; sim(a,b q ) is the similarity between sub-image a and the qth group of sub-images;
[0034] Among them, sub-image a comes from Q, b q Derived from Q-1, Q-1 represents the total number of sub-images in the set that does not contain sub-image a, Table pair ;
[0035] Based on the above formula, each group of sub-images in the set is taken as a, and the above formula is applied to calculate their comprehensive differences, and no less than two groups of sub-images with the largest comprehensive differences are selected and recorded as index image blocks.
[0036] Furthermore, the retrieval layer includes an upload module, a multiplexing module and a retrieval module, the upload module is used to upload a group of image data as a similar image retrieval target, the multiplexing module is used to receive the image data uploaded by the upload module, extract feature image blocks from the image data, and the retrieval module is used to receive the feature image blocks extracted by the multiplexing module, perform similarity comparison between the feature image blocks and each index image block in the index layer, and determine a group of image data with the highest similarity to the index image blocks;
[0037] After receiving the image data, the multiplexing module processes the image data in the same logic as the extraction module, and the multiplexing module extracts no less than two groups of feature image blocks.
[0038] Furthermore, when determining a group of image data with the highest similarity to the index image block, the retrieval module complies with:
[0039]
[0040] Where: SIMM is the similarity between the source image data of the feature image block and the corresponding image data of the index image block; c is the total number of feature image blocks; w is the total number of index image blocks derived from the same image data; sim(d p ,s t ) is the similarity between the p-th group of feature image blocks and the t-th group of index image blocks;
[0041] in, Table pair The average, Table pair The retrieval module calculates the similarity between the source image data of the feature image block and the image data corresponding to each index image block based on the above formula, and takes the image data corresponding to a group of index image blocks with the highest similarity as the similarity image data of the source image data of the feature image block, and outputs it;
[0042] In the image data output stage, the source path of the index image block of the image data is synchronously obtained and output together with the image data.
[0043] Furthermore, the control terminal is interactively connected to a receiving module via a wireless network, the receiving module is interactively connected to an extraction module and a marking module via a wireless network, the receiving module is interactively connected to a grouping module via a wireless network, the grouping module is interactively connected to a differentiation module and a storage module via a wireless network, the marking module is interactively connected to an upload module via a wireless network, and the upload module is interactively connected to a multiplexing module and a retrieval module via a wireless network.
[0044] Compared with the known public technology, the technical solution provided by the present invention has the following beneficial effects:
[0045] The present invention provides an image retrieval system based on similar images. During operation, the system manages the differentiated storage of image data so that the location where the image data is stored has logical similar image retrieval conditions when applied to similar image retrieval, thereby ensuring that the similar image retrieval process rate is more balanced and stable.
[0046] At the same time, in the specific similar image retrieval stage, the method of image segmentation and feature area image extraction is used to limit the similar image retrieval process to the extracted feature area image. Since the information contained in the feature area image is representative and smaller than the original image, the efficiency of the similar image retrieval process is improved. The similarity between the retrieved similar images and the retrieval target image in their respective feature positions is more accurate than the existing similar image retrieval technology. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0048] Figure 1 It is a structural diagram of an image retrieval system based on similar images;
[0049] Figure 2 This is a schematic diagram of the system storage layer image data storage logic in the present invention. DETAILED DESCRIPTION
[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0051] The present invention will be further described below in conjunction with the embodiments.
[0052] Embodiment 1:
[0053] An image retrieval system based on similar images in this embodiment, such as Figure 1 As shown, it includes: a control terminal, a storage layer, an index layer and a retrieval layer;
[0054] The control terminal is the main control terminal of the system and is used to issue execution commands;
[0055] The image data to be stored are uploaded through the storage layer and stored separately. The image data in each differentiated storage interval is differentiated again in the differentiated storage interval based on the image resolution. The index layer traverses the image data stored in each differentiated storage interval in the storage layer, extracts index image blocks on each group of image data, and marks the source path for each extracted index image block. The retrieval layer uploads a group of image data as a similar image retrieval target, extracts feature image blocks based on the index image block extraction logic in the similar image retrieval target, searches for similar index image blocks in the index image blocks extracted in the similar image retrieval target through the feature image blocks extracted, and simultaneously obtains the source path of the found index image block, determines the image data corresponding to the index image block based on the source path, and outputs the determined image data as the similar image data of the similar image retrieval target;
[0056] The storage layer includes a storage module, a distinguishing module, and a grouping module. The storage module is used to upload image data and store the uploaded image data. The distinguishing module is used to traverse the image data stored in the storage module, read the image data names, and build a distinguishing storage interval equal to the total amount of image names based on the total amount of different image data names in the storage module, and store image data with the same image name in the same distinguishing storage interval. The grouping module is used to identify the resolution of each image data stored in each distinguishing storage interval in the storage module, and perform a distinguishing operation on each image data stored in the distinguishing storage interval again according to the image data resolution.
[0057] Each set of image data uploaded in the storage module is labeled with a name, including: people, animals, scenery, objects, natural scenery, and weather;
[0058] The index layer includes a receiving module, an extraction module and a marking module. The receiving module is used to continuously receive the image data stored in the storage layer. The extraction module is used to synchronously obtain the image data received by the receiving module and extract the index image blocks from the image data. The marking module is used to continuously receive the index image blocks extracted by the extraction module and mark the source path of the index image blocks.
[0059] When the extraction module extracts the index image block from the image data, it first performs segmentation processing on the image data, and further performs the index image block extraction operation using the sub-images obtained by segmentation;
[0060] The image data segmentation process is subject to:
[0061]
[0062] Where: λ is the complexity of the image data; M and N are the number of rows and columns of the image data; δ x f(i,j) is the first-order difference of pixel (i,j) in the x direction; δ y f(i,j) is the first-order difference of pixel (i,j) in the y direction; E h is the horizontal symmetric error of the image data; E v is the vertical symmetric error of the image data; g is the image data segmentation parameter; γ is the adjustment factor;
[0063] The greater the complexity of the image data, the more information the image data contains. Conversely, the less information the image data contains. The image data segmentation parameter g is rounded up using the rounding method. After the image data segmentation parameter g is obtained, the number of image data segments is g. 2 , and the size of each segmented sub-image is equal;
[0064] The horizontal symmetry error of the image data and the vertical symmetry error of the image data are calculated by the following formula:
[0065]
[0066] In the formula: |I(x,y)-I h (x, y) is the original image I and the horizontal symmetric image I h The absolute value of the difference in pixel values at (x, y); |I(x, y)-I v (x,y) is the original image I and the vertical symmetric image I v The absolute value of the difference in pixel values at (x, y);
[0067] Among them, the horizontally symmetric image I h That is, the image obtained by flipping the original image along the horizontal central axis, the vertically symmetrical image Iv That is, the image obtained by flipping the original image along the vertical central axis;
[0068] The adjustment factor γ≥1, and follows the setting logic that the higher the image data resolution, the larger the adjustment factor, and vice versa;
[0069]
[0070] Where: f(i, j) is the pixel value of the image data at the coordinate (i, j);
[0071] In the stage of extracting index image blocks, the extraction module takes all sub-images obtained by segmenting the image data as a set, and extracts sub-images from the set as index image blocks:
[0072]
[0073] Where: DIFF a is the comprehensive difference between sub-image a and other sub-images in the set; Q is the total number of sub-images in the set; sim(a,b q ) is the similarity between sub-image a and the qth group of sub-images;
[0074] Among them, sub-image a comes from Q, b q Derived from Q-1, Q-1 represents the total number of sub-images in the set that does not contain sub-image a, Table pair ;
[0075] Based on the above formula, each group of sub-images in the set is taken as a, and the above formula is applied to calculate their comprehensive differences, and no less than two groups of sub-images with the largest comprehensive differences are selected and recorded as index image blocks;
[0076] The retrieval layer includes an upload module, a multiplexing module and a retrieval module. The upload module is used to upload a group of image data as a similar image retrieval target. The multiplexing module is used to receive the image data uploaded by the upload module and extract feature image blocks from the image data. The retrieval module is used to receive the feature image blocks extracted by the multiplexing module, compare the feature image blocks with each index image block in the index layer for similarity, and determine a group of image data with the highest similarity to the index image blocks.
[0077] After receiving the image data, the multiplexing module processes the image data in the same logic as the extraction module, and the multiplexing module extracts no less than two groups of feature image blocks;
[0078] The control terminal is interactively connected to a receiving module via a wireless network, the receiving module is interactively connected to an extraction module and a marking module via a wireless network, the receiving module is interactively connected to a grouping module via a wireless network, the grouping module is interactively connected to a differentiation module and a storage module via a wireless network, the marking module is interactively connected to an upload module via a wireless network, and the upload module is interactively connected to a multiplexing module and a retrieval module via a wireless network.
[0079] In this embodiment, the control terminal controls the operation of the control system, the storage module uploads the image data, and stores the uploaded image data. The distinguishing module synchronously traverses the image data stored in the storage module, reads the image data name, and constructs a distinguishing storage interval equal to the total amount of the image name based on the total amount of different image data names in the storage module, and stores the image data with the same image name in the same distinguishing storage interval. The grouping module further identifies the resolution of each image data stored in each distinguishing storage interval in the storage module, and distinguishes each image data stored in the distinguishing storage interval again according to the image data resolution. The receiving module performs a post-operation continuous receiving operation. The image data stored in the storage layer is then synchronously acquired by the extraction module from the image data received by the receiving module, and the index image blocks are extracted from the image data. The marking module synchronously and continuously receives the index image blocks extracted by the extraction module, and the source path of the index image blocks is marked. Then a group of image data is uploaded by the upload module as a similar image retrieval target. The reuse module receives the image data uploaded by the upload module in real time, extracts feature image blocks from the image data, and finally receives the feature image blocks extracted by the reuse module through the retrieval module, and compares the feature image blocks with each index image block in the index layer for similarity, and determines a group of image data with the highest similarity to the index image blocks.
[0080] Through the operation of the system in the above embodiment, different from the existing similar image retrieval technology, a retrieval logic with more balanced and stable retrieval rate and more efficient retrieval efficiency is provided for similar image retrieval;
[0081] See also Figure 2 As shown, it further shows the storage intervals distinguished in the storage module in the storage layer and the image data subsets further distinguished in the storage intervals. Based on the storage logic, the size of the image data in each subset is made as equal as possible, and each subset is further used as a single-run processing target, thereby ultimately achieving a balanced and stable retrieval rate effect.
[0082] Embodiment 2:
[0083] In terms of specific implementation, based on Example 1, this example refers to Figure 1 The image retrieval system based on similar images in Example 1 is further described in detail:
[0084] In the grouping module operation stage, after identifying the resolution of each image data in each partitioned storage interval, the image data stored in each partitioned storage interval is arranged in descending or ascending order based on the resolution of the image data, and the first and last groups of image data in the image data arranged in each partitioned storage interval are continuously selected as a subset;
[0085] When the number of image data stored in the distinguishing storage interval is odd, a group of image data in the middle of the image data arranged in descending or ascending order based on the resolution of the image data in the distinguishing storage interval is taken as a subset;
[0086] In the receiving module operation phase, each subset in each storage interval of the storage module is taken as a receiving target, and the receiving module continuously receives the image data stored in the storage layer, that is, continuously receives the subset in the storage module;
[0087] When the marking module marks the source path of the index image block, the location of the subset where the source image data of the index image block is located and the location of the partitioned storage interval where the subset is located are used as marking content to perform the marking operation of the source path of the index image block.
[0088] In this embodiment, the above-mentioned setting is used to further limit the storage logic for distinguishing image data in the storage module in the system in Example 1, and the index image blocks extracted from the image data are marked to facilitate tracing and determining the source image data of the index image blocks when the system is further running.
[0089] Embodiment 3:
[0090] In terms of specific implementation, based on Example 1, this example refers to Figure 1 The image retrieval system based on similar images in Example 1 is further described in detail:
[0091] When the retrieval module determines a set of image data with the highest similarity to the index image block, it follows:
[0092]
[0093] Where: SIMM is the similarity between the source image data of the feature image block and the corresponding image data of the index image block; c is the total number of feature image blocks; w is the total number of index image blocks derived from the same image data; sim(d p ,s t ) is the similarity between the p-th group of feature image blocks and the t-th group of index image blocks;
[0094] in, Table pair The average, Table pair The retrieval module calculates the similarity between the source image data of the feature image block and the image data corresponding to each index image block based on the above formula, and takes the image data corresponding to a group of index image blocks with the highest similarity as the similarity image data of the source image data of the feature image block, and outputs it;
[0095] In the image data output stage, the source path of the index image block of the image data is synchronously obtained and output together with the image data.
[0096] In this embodiment, the above logic formula provides a logical basis for the final determination of similar image data in the system retrieval layer in Embodiment 1, ensuring that the retrieval layer stably outputs similar image data of similar image retrieval targets.
[0097] In summary, during the operation of the system in the above embodiment, the image data is differentiated, stored and managed so that the location where the image data is stored has logical similar image retrieval conditions when applied to similar image retrieval, thereby ensuring that the rate of the similar image retrieval process is more balanced and stable. At the same time, in the specific similar image retrieval stage, the image segmentation and feature area image extraction methods are used to limit the similar image retrieval process to the extracted feature area image. Since the information contained in the feature area image is representative and smaller than the original image, the efficiency of the similar image retrieval process is improved. In addition, the similarity between the retrieved similar images and the retrieval target image at their respective feature positions is more accurate than the existing similar image retrieval technology.
[0098] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An image retrieval system based on similar images, characterized in that: include: Control terminal, storage layer, index layer and retrieval layer; The control terminal is the main control terminal of the system and is used to issue execution commands; The image data to be stored are uploaded through the storage layer and stored separately. The image data in each differentiated storage interval is differentiated again in the differentiated storage interval based on the image resolution. The index layer traverses the image data stored in each differentiated storage interval in the storage layer, extracts index image blocks on each group of image data, and marks the source path for each extracted index image block. The retrieval layer uploads a group of image data as a similar image retrieval target, extracts feature image blocks based on the index image block extraction logic in the similar image retrieval target, searches for similar index image blocks in the index image blocks extracted in the similar image retrieval target through the feature image blocks extracted, and simultaneously obtains the source path of the found index image block, determines the image data corresponding to the index image block based on the source path, and outputs the determined image data as the similar image data of the similar image retrieval target; The index layer includes a receiving module, an extracting module and a marking module, wherein the receiving module is used to continuously receive the image data stored in the storage layer, the extracting module is used to synchronously obtain the image data received by the receiving module, extract the index image blocks from the image data, and the marking module is used to continuously receive the index image blocks extracted by the extracting module, and mark the source path of the index image blocks; When the extraction module extracts the index image block from the image data, it first performs segmentation processing on the image data, and further performs the index image block extraction operation using the sub-images obtained by segmentation; The image data segmentation process is subject to: Where: λ is the complexity of the image data; M and N are the number of rows and columns of the image data; δ x f(i,j) is the first-order difference of pixel (i,j) in the x direction; δ y f(i,j) is the first-order difference of pixel (i,j) in the y direction; E h is the horizontal symmetric error of the image data; E v is the vertical symmetric error of the image data; g is the image data segmentation parameter; γ is the adjustment factor; The greater the complexity of the image data, the more information the image data contains. Conversely, the less information the image data contains. The image data segmentation parameter g is rounded up using the rounding method. After the image data segmentation parameter g is obtained, the number of image data segments is g. 2 , and the size of each segmented sub-image is equal.
2. The image retrieval system based on similar images according to claim 1, characterized in that: The storage layer includes a storage module, a distinguishing module, and a grouping module. The storage module is used to upload image data and store the uploaded image data. The distinguishing module is used to traverse the image data stored in the storage module, read the image data names, and based on the total amount of different image data names, construct a distinguishing storage interval equal to the total amount of image names in the storage module, and store image data with the same image name in the same distinguishing storage interval. The grouping module is used to identify the resolution of each image data stored in each distinguishing storage interval in the storage module, and perform a distinguishing operation on each image data stored in the distinguishing storage interval again according to the image data resolution. Each set of image data uploaded in the storage module is labeled with a name, including: people, animals, scenery, objects, natural scenery, and weather.
3. The image retrieval system based on similar images according to claim 2, characterized in that: During the grouping module operation phase, after identifying the resolution of each image data in each partitioned storage interval, the image data stored in each partitioned storage interval is sorted in descending or ascending order based on the resolution of the image data, and the first and last groups of image data in the image data continuously sorted in each partitioned storage interval are selected as a subset; When the number of image data stored in the distinguished storage interval is odd, a middle group of image data in the distinguished storage interval arranged in descending or ascending order based on the image data resolution is taken as a subset.
4. The image retrieval system based on similar images according to claim 1, characterized in that: During the operation phase of the receiving module, each subset in each storage interval of the storage module is taken as a receiving target, and the receiving module continuously receives the image data stored in the storage layer, that is, continuously receives the subset in the storage module; When the marking module marks the source path of the index image block, the subset where the source image data of the index image block is located and the location of the differentiated storage interval where the subset is located are used as marking content to perform the marking operation of the source path of the index image block.
5. The image retrieval system based on similar images according to claim 1, characterized in that: The horizontal symmetric error of the image data and the vertical symmetric error of the image data are obtained by the following formula: In the formula: |I(x,y)-I h (x, y) is the original image I and the horizontal symmetric image I h The absolute value of the difference in pixel values at (x, y); |I(x, y)-I v (x,y) is the original image I and the vertical symmetric image I v The absolute value of the difference in pixel values at (x, y); Among them, the horizontally symmetric image I h That is, the image obtained by flipping the original image along the horizontal central axis, the vertically symmetrical image I v That is, the image obtained by flipping the original image along the vertical central axis.
6. The image retrieval system based on similar images according to claim 1, characterized in that: The adjustment factor γ≥1, and follows the setting logic that the higher the image data resolution, the larger the adjustment factor, and vice versa; Where: f(i,j) is the pixel value of the image data at the coordinate position (i,j).
7. The image retrieval system based on similar images according to claim 1, characterized in that: In the stage of extracting index image blocks, the extraction module takes all sub-images obtained by segmenting the image data as a set, and extracts sub-images from the set as index image blocks: Where: DIFF a is the comprehensive difference between sub-image a and other sub-images in the set; Q is the total number of sub-images in the set; sim(a,b q ) is the similarity between sub-image a and the qth group of sub-images; Among them, sub-image a comes from Q, b q Derived from Q-1, Q-1 represents the total number of sub-images in the set that does not contain sub-image a, Table pair ; Based on the above formula, each group of sub-images in the set is taken as a, and the above formula is applied to calculate their comprehensive differences, and no less than two groups of sub-images with the largest comprehensive differences are selected and recorded as index image blocks.
8. The image retrieval system based on similar images according to claim 1, characterized in that: The retrieval layer includes an upload module, a multiplexing module and a retrieval module. The upload module is used to upload a group of image data as a similar image retrieval target. The multiplexing module is used to receive the image data uploaded by the upload module and extract feature image blocks from the image data. The retrieval module is used to receive the feature image blocks extracted by the multiplexing module, compare the feature image blocks with each index image block in the index layer for similarity, and determine a group of image data with the highest similarity to the index image blocks. After receiving the image data, the multiplexing module processes the image data in the same logic as the extraction module, and the multiplexing module extracts no less than two groups of feature image blocks.
9. The image retrieval system based on similar images according to claim 8, characterized in that: When determining a set of image data with the highest similarity to the index image block, the retrieval module complies with: Where: SIMM is the similarity between the source image data of the feature image block and the image data corresponding to the index image block; c is the total amount of feature image blocks; w is the total number of index image blocks from the same image data; sim(d p ,s t ) is the similarity between the p-th group of feature image blocks and the t-th group of index image blocks; in, Table pair The average, Table pair The retrieval module calculates the similarity between the source image data of the feature image block and the image data corresponding to each index image block based on the above formula, and takes the image data corresponding to a group of index image blocks with the highest similarity as the similarity image data of the source image data of the feature image block, and outputs it; In the image data output stage, the source path of the index image block of the image data is synchronously obtained and output together with the image data.
10. The image retrieval system based on similar images according to claim 1, characterized in that: The control terminal is interactively connected to a receiving module via a wireless network, the receiving module is interactively connected to an extraction module and a marking module via a wireless network, the receiving module is interactively connected to a grouping module via a wireless network, the grouping module is interactively connected to a differentiation module and a storage module via a wireless network, the marking module is interactively connected to an upload module via a wireless network, and the upload module is interactively connected to a multiplexing module and a retrieval module via a wireless network.
Citation Information
Patent Citations
Multi-feature VP-tree index-based remote sensing image retrieval method and multi-feature VP-tree index-based remote sensing image retrieval device
CN104239551A
Determination method and device of segmentation mode, display method and equipment, and storage medium
CN110782463A
Image recognition system for camera shooting
CN117994489A
Image retrieving system, image classifying system, image retrieving program, image classifying program, image retrieving method and image classifying method
CN1573742A
Similar image retrieving method, similar image retrieving system, similar image retrieving program, and recording medium
JP2006338313A
