CN113313126A - Method, computing device, and computer storage medium for image recognition - Google Patents
Method, computing device, and computer storage medium for image recognition Download PDFInfo
- Publication number
- CN113313126A CN113313126A CN202110482595.9A CN202110482595A CN113313126A CN 113313126 A CN113313126 A CN 113313126A CN 202110482595 A CN202110482595 A CN 202110482595A CN 113313126 A CN113313126 A CN 113313126A
- Authority
- CN
- China
- Prior art keywords
- image
- data set
- image data
- texture
- color
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 103
- 230000004927 fusion Effects 0.000 claims abstract description 36
- 238000010219 correlation analysis Methods 0.000 claims abstract description 25
- 238000000605 extraction Methods 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims description 120
- 239000013598 vector Substances 0.000 claims description 101
- 238000004364 calculation method Methods 0.000 claims description 16
- 230000009467 reduction Effects 0.000 claims description 14
- 238000007499 fusion processing Methods 0.000 claims description 13
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 abstract description 10
- 230000000007 visual effect Effects 0.000 abstract description 5
- 238000012163 sequencing technique Methods 0.000 abstract description 4
- 238000005259 measurement Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 15
- 238000012545 processing Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 8
- 239000000284 extract Substances 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 101100043112 Homo sapiens SERPINB3 gene Proteins 0.000 description 3
- 102100036383 Serpin B3 Human genes 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000007500 overflow downdraw method Methods 0.000 description 2
- SLVOVFVZZFUEAS-UHFFFAOYSA-N 2-[2-[2-[bis(carboxymethyl)amino]ethoxy]ethyl-(carboxymethyl)amino]acetic acid Chemical compound OC(=O)CN(CC(O)=O)CCOCCN(CC(O)=O)CC(O)=O SLVOVFVZZFUEAS-UHFFFAOYSA-N 0.000 description 1
- 241000124008 Mammalia Species 0.000 description 1
- 230000031018 biological processes and functions Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000002207 retinal effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method for image recognition, which comprises the steps of firstly extracting color features and texture features of an image, and then performing feature extraction by using a CCA family correlation analysis algorithm; secondly, sparse representation of features is carried out by utilizing a depth sparse self-coding operator, so that the overall data volume of the image is reduced; and completing the fusion of the two visual angle characteristics in a parallel or serial mode; and finally, calculating the similarity of the retrieval images by using a distance measurement formula, and selecting the first images in the sequencing result as retrieval results. The invention solves the problems of low image recognition rate and high operation overhead in the prior art.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to a method, a computing device, and a computer storage medium for image recognition.
Background
In the prior art, image retrieval methods are mainly divided into two types: one is image retrieval at a text level, and the other is image retrieval at a content level. The image retrieval of the text level depends on the word expression, if the description keyword error of the image can cause the deviation of the retrieval result, when facing a large amount of network images, the image retrieval method has the defects of huge workload of manual labeling and one-sided description. And the retrieval of the content image focuses on the extraction of the image content, so that the method has the advantages of intuition and convenience. However, because of many image features and large sample size, the content image retrieval process is often designed to be complicated, so that the current mainstream image search still uses the text retrieval method as the main part of the system.
For image retrieval, how to reasonably extract image features and reduce data redundancy is a very critical problem. The image features are of various types, such as global features, color features and texture features, wherein the global features are overall quantitative representation of the image, the color features are data which can be directly observed by human eyes and are not affected by image deformation, the texture features are information which is formed by image pixel points according to a certain arrangement rule, and the selection of different features can directly cause the change of a retrieval result, so that the use of which feature extraction method has important significance for image retrieval.
In practical application, the current ideal feature extraction methods include the following two methods:
the Ncut technology, normalized cut, is one of the image segmentation methods based on graph theory, firstly divides the graph into regions with the same characteristics, then applies recursive algorithm and takes the set norm as the iteration stop condition, although the marginalization deviation can be reduced, the recognition rate is still not high, the calculation cost is large and the iteration speed is low.
The Gabor filtering method, which was first described in Gabor D's argument and later developed into 2D, can simultaneously acquire uncertainty in time and frequency domains, and is consistent with the receiving field of retinal nerve cells of mammals, and thus is applied to many fields of computer vision, especially to feature extraction using Gabor filters, but there are many problems in practical application, such as large computational overhead and storage burden.
Therefore, a fast and accurate similar image retrieval method is needed to solve the problems of low image recognition rate and high operation cost in the prior art.
Disclosure of Invention
In view of the technical problem of the prior art that the complexity for searching similar images is large, the embodiments of the present invention provide a method, a computing device and a computer storage medium for image recognition.
In a first aspect, an embodiment of the present invention provides a method for similar image retrieval, including: acquiring an image to be identified and an image data set; extracting color features and texture features of the image to be recognized to generate color feature vectors and texture feature vectors of the image to be recognized; performing CCA (Canonical correlation analysis) family correlation analysis calculation on the image data set aiming at the color feature vector and the texture feature vector of the image to be identified so as to generate a color feature matrix and a texture feature matrix of the image data set; respectively carrying out fusion processing on the color characteristic vector and the texture characteristic vector of the image to be identified and the color characteristic matrix and the texture characteristic matrix of the image data set by adopting a parallel or serial connection method to obtain a fusion characteristic vector of the image to be identified and a global characteristic matrix of the image data set; and calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be identified according to the global characteristic matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining a retrieval result which is most similar to the image to be identified.
Further, the acquiring the image to be recognized and the image dataset includes: a user inputs an image to be recognized and an image data set; or receiving the image to be identified and the image data set through a preset data transmission interface; or the user inputs one of the image to be identified and the image data set, and the other one is received and obtained through a preset data transmission interface.
Further, the extracting color features and texture features of the image to be recognized to generate color feature vectors and texture feature vectors of the image to be recognized includes: and respectively extracting the color features and the texture features of the image to be identified by utilizing a color histogram and a Gabor feature extraction method so as to generate the color feature vector and the texture feature vector of the image to be identified.
Further, after the performing CCA family correlation analysis and calculation on the image data set aiming at the color feature vector and the texture feature vector of the image to be recognized to generate a color feature matrix and a texture feature matrix of the image data set, before performing fusion processing on the color feature vector and the texture feature vector of the image to be recognized and the color feature matrix and the texture feature matrix of the image data set respectively by using a parallel or serial method to obtain a fusion feature vector of the image to be recognized and a global feature matrix of the image data set, the method further includes: and respectively carrying out dimension reduction treatment on the color characteristic matrix and the texture characteristic matrix of the image data set by adopting a standard correlation analysis method to obtain the effective color characteristic matrix and texture characteristic matrix of the image data set after dimension reduction.
Further, after the performing fusion processing on the color feature vector and the texture feature vector of the image to be recognized and the color feature matrix and the texture feature matrix of the image data set by using a parallel or serial method to obtain the fusion feature vector of the image to be recognized and the global feature matrix of the image data set, before calculating the correlation between the fusion features of the image data set and the image to be recognized according to the global feature matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining a retrieval result most similar to the image to be recognized, the method further includes: and performing multilayer feature reconstruction on the global feature matrix of the image data set by using a depth extreme learning machine to obtain the global feature matrix of the image data set in sparse representation.
Further, after the performing CCA family correlation analysis and calculation on the image data set aiming at the color feature vector and the texture feature vector of the image to be recognized to generate a color feature matrix and a texture feature matrix of the image data set, before performing fusion processing on the color feature vector and the texture feature vector of the image to be recognized and the color feature matrix and the texture feature matrix of the image data set respectively by using a parallel or serial method to obtain a fusion feature vector of the image to be recognized and a global feature matrix of the image data set, the method further includes: and performing multilayer feature reconstruction on the color feature matrix and the texture matrix of the image data set by using a deep extreme learning machine to obtain the color feature matrix and the texture feature matrix of the image data set which are expressed sparsely.
Further, the calculating, according to the global feature matrix of the image data set, a correlation degree of a fusion feature vector between the image data set and the image to be recognized, performing descending order arrangement on the images in the image data set, and determining a retrieval result most similar to the image to be recognized includes: and calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be recognized according to the global characteristic matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining the image arranged at the first position as the recognition result most similar to the image to be recognized to be output.
Further, the calculating, according to the global feature matrix of the image data set, a correlation degree of a fusion feature vector between the image data set and the image to be recognized, performing descending order arrangement on the images in the image data set, and determining a retrieval result most similar to the image to be recognized includes: and calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be identified according to the global characteristic matrix of the image data set, responding to the calculation result of the correlation degree and a preset retrieval amount threshold value, performing descending arrangement on the images in the image data set according to the correlation degree, and outputting the images in the preset retrieval amount threshold value which is ranked in the front as the retrieval result.
In a second aspect, an embodiment of the present invention provides a computing device, which includes a memory, a processor, and a program stored in the memory and executable on the processor, where the processor implements the steps of the method according to any one of the first aspect to the seventh possible implementation manner of the first aspect when executing the program.
In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the method according to any one of the first aspect to the sixth possible implementation manner of the first aspect.
As can be seen from the above, one or more technical solutions provided in the embodiments of the present invention at least achieve the following technical effects or advantages:
(1) the invention adopts the image retrieval method of the fusion of the two-view-angle characteristics of the depth sparse self-coding operator aiming at the color characteristics and the texture characteristics, thereby avoiding the inaccuracy of image identification by adopting single characteristics in the prior art, obtaining more accurate identification results and effectively improving the identification rate of image retrieval;
(2) the method firstly extracts color features and texture features of the image, and then extracts the features by using a CCA (Canonical correlation analysis) family algorithm; then, carrying out sparse representation on the features by using a DSAE (Doherty Sommers architectures Engineers depth sparse self-coding operator), thereby reducing the overall data volume of the image and completing the feature fusion of two visual angles in a parallel or serial mode; and then calculating the correlation of the retrieval images, and selecting the images at the front in the sequencing result as retrieval results. Therefore, the method avoids the operation complexity increased by multi-level iteration in the prior art, solves the technical problems of high iteration overhead and low operation efficiency of massive image retrieval, improves the calculation speed and greatly reduces the occupation of storage space.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a flowchart of a method for image recognition according to an embodiment of the present invention;
FIG. 2 is a flowchart of a method for image recognition according to an embodiment of the present invention;
FIG. 3 is a flowchart of a method for image recognition according to an embodiment of the present invention;
FIG. 4 is a flowchart of a method for image recognition according to an embodiment of the present invention;
FIG. 5 is a block diagram of a process for an image recognition system according to an embodiment of the present invention;
FIG. 6 is a block diagram of a process for an image recognition system according to an embodiment of the present invention;
FIG. 7 is a block diagram of a process for an image recognition system according to an embodiment of the present invention;
FIG. 8 is a block diagram of a process for an image recognition system according to an embodiment of the present invention;
FIG. 9 is a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention;
FIG. 10 is a schematic structural diagram of a computer device according to an embodiment of the present invention;
FIG. 11 is a diagram illustrating an exemplary process for processing trademark image features according to an embodiment of the present invention;
fig. 12 is a diagram illustrating an exemplary process of matching trademark identification features according to an embodiment of the present invention.
Detailed Description
In view of the technical problems of low image recognition rate and high operation cost in the prior art, the embodiment of the invention provides a method for recognizing an image based on a depth limit learning machine, and the general idea is as follows:
for image retrieval, how to reasonably extract image features and reduce data redundancy is a very critical problem, and a deep extreme learning machine can process high-dimensional data, has a fast iteration speed and a strong feature extraction capability, so that the deep extreme learning machine is applied to the image retrieval. The image features are of various types, such as global features, color features and texture features, wherein the global features are expressed in an integral manner of the image, the color features are data which can be directly observed by human eyes and are not affected by image deformation, and the texture features are information which is formed by image pixel points according to a certain arrangement rule, and different features are selected to directly cause the retrieval result to change. In practical application, the retrieval effect obtained by only selecting a single image feature is not ideal, so that the feature fusion method has important significance for image retrieval. Therefore, in order to improve the feature processing process in image retrieval, the invention provides an image retrieval method based on two-view feature fusion of a depth sparse self-coding operator (Doherty Sommers technologies Engineers, DSAE) from the aspect of features, firstly, extracting color features and texture features of an image, and then extracting the features by using a CCA family algorithm; then, the DSAE is used for sparse representation of the features, so that the overall data volume of the image is reduced, and the two-view feature fusion is completed in a parallel or serial mode; and then, calculating the similarity of the retrieval images by using a distance measurement formula, and selecting a plurality of previous images in the sequencing result as an output result. The method is characterized in that a new method is constructed based on the depth limit learning machine, the similar image discovery is realized, and the user is helped to automatically complete the image recognition. The method aims to improve the accuracy of image recognition, reduce the operation overhead through dimension reduction processing and improve the operation efficiency of image recognition while improving the image recognition rate.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
An embodiment of the present invention provides a method for image recognition, please refer to fig. 1, which includes the following steps:
first, step S101 is performed: an image to be identified and an image dataset are acquired.
Specifically, the image to be retrieved and the image data set may adopt the following embodiments:
a1, inputting an image to be retrieved and an image data set by a user; or
A2, receiving an image to be retrieved and an image data set through a preset data transmission interface; or
A3, inputting one of the image to be retrieved and the image data set by the user, and receiving the other one through the preset data transmission interface.
After step S101, step S102 is then performed: and extracting the color features and the texture features of the image to be recognized to generate the color feature vector and the texture feature vector of the image to be recognized.
It should be noted that there are many types of image features, such as global features, color features, and texture features, where the global features are overall quantized representations of an image, the color features are data that can be directly observed by human eyes and are not affected by image deformation, and the texture features are information that is composed of image pixels according to a certain arrangement rule, and selecting different features directly causes a change in a retrieval result. In practical application, the retrieval effect obtained by only selecting a single image feature is not ideal, so that the feature fusion method has significance for image retrieval, the identification rate of the image can be effectively improved, and the identification of the image is more accurate.
In order to implement feature fusion, preferably, the extracting color features and texture features of the image to be recognized to generate color feature vectors and texture feature vectors of the image to be recognized includes:
and respectively extracting the color features and the texture features of the image to be identified by utilizing a color histogram and a Gabor feature extraction method so as to generate the color feature vector and the texture feature vector of the image to be identified.
The color histogram feature extraction is used to reflect the composition distribution of the image colors, i.e., the probability of the appearance of each color. The histogram is used as a simple and effective feature descriptor based on statistical characteristics and is widely used in the field of computer vision. Its advantages are two aspects: firstly, for any image area, the extraction of the histogram features is simple and convenient; secondly, the histogram represents the statistical characteristics of the image region, can effectively represent the multi-modal feature distribution, and has a certain rotation invariance. The method comprises the steps of firstly obtaining a color histogram by stripping three components of a color space, and then finding that the color histogram of an image is not changed greatly after the image is subjected to rotation transformation, scaling transformation and fuzzy transformation by observing experimental data, namely the image histogram is insensitive to the physical transformation of the image. In the Gabor feature extraction method, the Gabor transform belongs to windowed fourier transform, and the Gabor function can extract related features in different scales and different directions of a frequency domain. In addition, the Gabor function is similar to the biological action of human eyes, so that the Gabor function is often used for texture recognition and achieves better effect. Therefore, the method for extracting the color features and the texture features and extracting the color histogram and the Gabor features can obtain the description data which can directly describe the visual features of the image in the image pixels and is applied to measuring and comparing the global difference of the two images.
After step S102, step S103 is then performed: and performing CCA family correlation analysis calculation on the image data set aiming at the color characteristic vector and the texture characteristic vector of the image to be identified so as to generate a color characteristic matrix and a texture characteristic matrix of the image data set.
It should be noted that, in the present invention, a typical correlation analysis (CCA) is a classical two-view dimensionality reduction method, which mainly focuses on the maximum correlation between two sets of random variables, and the basic idea is to find a set of projection vectors to maximize the correlation coefficient thereof, and use the projection vectors to replace the relationship between the original two sets of random variables. Assuming that the two sets of features of sample X are X1 and X2, respectively, X1 has a dimension p1, X2 has a dimension p2, and p1 ≦ p 2.
The correlation coefficient can be expressed as:
the larger Corr (u, v) is expected to be the better, so the optimization function is expressed as equation (3), which can be solved using the lagrangian method.
Semi-Supervised Canonical Correlation Analysis (SCCA) is the product of Semi-supervised learning in conjunction with traditional CCA. For a given two sets of random vectors, the CCA expects to find the maximum correlation between samples, ensuring that the correlation coefficient is the largest, and thus obtain the projected feature vector. However, CCA does not take into account supervised information in the samples and known information cannot be effectively utilized. And the SCCA adds a small amount of prior constraint information on the basis of the CCA, and better constrains the correlation among the feature vectors so as to obtain better feature vector representation.
Rank Canonical Correlation Analysis (RCCA) adds correlation constraints to CCA, overcomes the disadvantage that other dimensionality reduction methods ignore supervised information, can train labeled and unlabeled data simultaneously, and retains the superiority of low-dimensional features.
The Local Discriminant Canonical Correlation Analysis (LDCCA) is an algorithm which is provided on the basis of the CCA and can solve a large number of nonlinear problems, the aim of the algorithm is to search a group of projections to enable the correlation among similar k adjacent data to be maximum and the correlation among different k adjacent data to be minimum, and the defect that only paired sample correlation can be used is overcome through the class information of samples.
The Local Preservation Canonical Correlation Analysis (LPCCA) converts a global non-linear problem into a local linear problem, introduces local structure information into the CCA, calculates a correlation problem in each small neighborhood, and optimizes and calculates to obtain a group of projection vectors. The LPCCA can better reserve the popular information of the initial training sample, the global non-linearity problem is solved through a local linear method, and the low-dimensional features of the sample are effectively extracted.
Typical correlation analysis (DMPCCA) for distinguishing Minimum class local retention combines local structure information and a global distinguishing rule, is similar to the CCA, two projection vectors with the maximum correlation are searched, the distinguishing capability between the local information in the class and the class is ensured to be maximum, and the DMPCCA well retains the local structure of a sample, so that the distinguishing information is effectively reflected.
Of course, the present invention may also adopt an SCCA, RCCA, LDCCA, LPCCA, or DMPCCA algorithm to perform CCA family correlation analysis calculation on the image data set with respect to the color feature vector and the texture feature vector of the image to be recognized, so as to generate a color feature matrix and a texture feature matrix of the image data set.
After step S103, step S104 is then performed: and respectively carrying out fusion processing on the color characteristic vector and the texture characteristic vector of the image to be identified and the color characteristic matrix and the texture characteristic matrix of the image data set by adopting a parallel or serial connection method to obtain the fusion characteristic vector of the image to be identified and the global characteristic matrix of the image data set.
It should be noted that, if the color feature vector of the image to be recognized is Px and the texture feature vector is Py, and the color feature matrix of the image data set is Qx and the texture feature matrix is Qy, then obtaining the fusion feature vector Xs of the image to be recognized and the global feature matrix Qs of the image data set in a parallel or serial manner respectively is:
parallel connection: xs ═ Px + Py; qs is Qx + Qy
Series connection: xs ═ Px, Py ]; qs is [ Qx, Qy ].
After step S104, step S105 is executed to calculate the correlation of the fusion feature vector between the image data set and the image to be recognized according to the global feature matrix of the image data set, perform descending order arrangement on the images in the image data set, and determine the retrieval result most similar to the image to be recognized.
In specific implementation, the similarity calculation can be performed by using the Euclidean distance, the results are sorted, and the label of the first image is extracted and output as the retrieval result. In the present invention, the euclidean distance is used for the correlation measurement, and then the euclidean distance between the sample points x11, x 12., x1N and x21, x 22., x2N in the N-dimensional space is:
in order to reduce the overhead of the operation, in one embodiment, as shown in fig. 2, after step S103 and before step S104, the following step a is further included: and respectively carrying out dimension reduction on the color characteristic matrix and the texture characteristic matrix of the image data set by adopting a standard correlation analysis method to obtain the effective color characteristic matrix and texture characteristic matrix of the image data set after dimension reduction.
In this embodiment, after step a, step S104 is executed: and respectively carrying out fusion processing on the color characteristic vector and the texture characteristic vector of the image to be identified and the color characteristic matrix and the texture characteristic matrix of the image data set after dimension reduction by adopting a parallel connection or series connection method to obtain the fusion characteristic vector of the image to be identified and the global characteristic matrix of the image data set.
In order to further reduce the overhead of the operation, in an embodiment, as shown in fig. 3, after step S104 and before step S105, the following step B is further included: and carrying out multilayer feature reconstruction on the global feature matrix of the image data set by using a deep extreme learning machine to obtain the global feature matrix of the image data set in sparse representation.
In this embodiment, after step B, step S105 is executed: according to the sparsely represented global feature matrix of the image data set, calculating the correlation degree of the fusion feature vector of the image data set and the image to be recognized, performing descending order arrangement on the images in the image data set, and determining the retrieval result which is most similar to the image to be recognized.
In order to further optimize the image features and enhance the description capability of the extracted image features, in an embodiment, as shown in fig. 4, after step S103 and before step S104, a step C of performing multi-layer feature reconstruction on the color feature matrix and the texture matrix of the image data set by using a depth extreme learning machine to obtain a color feature matrix and a texture feature matrix of the image data set in a sparse representation is further included.
In this embodiment, after step C, step S104 is executed: and respectively carrying out fusion processing on the color characteristic vector and the texture characteristic vector of the image to be identified and the color characteristic matrix and the texture characteristic matrix of the image data set which are sparsely represented by adopting a parallel or serial method to obtain a fusion characteristic vector of the image to be identified and a global characteristic matrix of the image data set.
In any of the above embodiments, step S105) calculates the correlation between the image dataset and the fused feature vector of the image to be recognized according to the global feature matrix of the image dataset, performs descending order arrangement on the images in the image dataset, and determines the search result most similar to the image to be recognized, which specifically includes the following two execution cases:
calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be recognized according to the global characteristic matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining the image arranged at the first position as the recognition result most similar to the image to be recognized to be output; or
And calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be identified according to the global characteristic matrix of the image data set, responding to the calculation result of the correlation degree and a preset retrieval amount threshold value, arranging the images in the image data set in a descending order according to the correlation degree, and outputting the image in the preset retrieval amount threshold value which is arranged in the front order as a retrieval result. For example, if the preset search threshold is 12, the images in the image data set are sorted in descending order according to the degree of correlation, and the top 12 images sorted in the front are output as the search result.
In the specific implementation process, the method for image recognition provided by the embodiment of the invention can be applied to recognition of medical consumable images with irregular images, and the practical application of the method of the invention is further described below by taking medical consumable trademark recognition based on a depth extreme learning machine as an example:
firstly, extracting a quotient mark foreground from an input medical consumable image by adopting an Ncut technology and a saliency target detection model to form a candidate identification target; secondly, extracting the feature information of the trademark foreground image by using a color histogram and Gabor feature extraction method; then respectively reducing the dimension of the features of the two visual angles of the color and the texture by using a standard correlation analysis method and the like, taking the dimension after dimension reduction as a column and the number of the trademark foreground images as a row to obtain an effective color and texture feature matrix; fusing the color characteristic matrix and the texture characteristic matrix by using a parallel or serial method to obtain a global characteristic matrix; then, carrying out multi-layer feature reconstruction on the global features by using a deep extreme learning machine to obtain a sparsely represented global feature matrix; fig. 11 shows an exemplary process of processing the trademark image features, and the specific process of processing the features can be divided into four stages: in the first stage, a color histogram method and a Gabor filtering method are respectively used for extracting color features and texture features of original image data, so that color features A and texture features A are obtained; in the second stage, the two-view-angle algorithms such as CCA (clear channel assessment) are used for respectively reducing the dimensions of the features of the two view angles of the color and the texture to obtain a color feature B and a texture feature B; in the third stage, the color features and the texture features are fused by using a parallel or serial method, so that a processed global feature C is obtained; in the fourth stage, a deep extreme learning machine is used for carrying out multilayer reconstruction calculation on the global feature C, and finally the global feature D in sparse representation is obtained, so that feature processing of the image is completed.
Finally, measuring the similarity between the images in a proper mode, and then quantitatively analyzing the correlation between the query images and the feature database; then, according to the correlation degree of the feature database data and the query image features, images in the image database are arranged in a descending order, and finally, the first images in the sequence are returned in the result and output as the retrieval result. An example of a specific brand identification feature matching process is shown in fig. 12. A large number of experiments prove that the method is very effective and can obtain a relatively accurate identification result.
Based on the same inventive concept, an embodiment of the present invention provides an image recognition system, which is shown in fig. 5 and specifically includes:
an obtaining unit 501, configured to obtain an image to be identified and an image data set;
an extracting unit 502, configured to extract color features and texture features of the image to be recognized to generate color feature vectors and texture feature vectors of the image to be recognized;
a correlation analysis unit 503, configured to perform CCA family correlation analysis calculation on the image data set for the color feature vector and the texture feature vector of the image to be identified, so as to generate a color feature matrix and a texture feature matrix of the image data set;
a fusion processing unit 504, configured to perform fusion processing on the color feature vector and the texture feature vector of the image to be identified and the color feature matrix and the texture feature matrix of the image data set by using a parallel or serial method, respectively, to obtain a fusion feature vector of the image to be identified and a global feature matrix of the image data set;
and the output unit 505 is configured to calculate a correlation between the image data set and the fusion feature vector of the image to be recognized according to the global feature matrix of the image data set, perform descending order arrangement on the images in the image data set, and determine a retrieval result most similar to the image to be recognized.
In an embodiment, referring to fig. 6, an image recognition system according to an embodiment of the present invention further includes:
and the dimension reduction processing unit 5A is configured to perform dimension reduction processing on the color feature matrix and the texture feature matrix of the image data set respectively by using a canonical correlation analysis method, so as to obtain the color feature matrix and the texture feature matrix of the image data set which are effective after dimension reduction.
In one embodiment, referring to fig. 7, an image recognition system according to an embodiment of the present invention further includes:
and the global feature reconstruction unit 5B is configured to perform multilayer feature reconstruction on the full local feature matrix of the image data set by using a depth extreme learning machine, so as to obtain a global feature matrix of the image data set, which is sparsely represented.
In an embodiment, referring to fig. 8, an image recognition system according to an embodiment of the present invention further includes:
and the single feature reconstruction unit 5C is configured to perform multi-layer feature reconstruction on the color feature matrix and the texture matrix of the image data set by using a deep extreme learning machine to obtain a color feature matrix and a texture feature matrix of the image data set, which are sparsely represented.
Based on the same inventive concept, an embodiment of the present invention provides a computer-readable storage medium 901, as described with reference to fig. 9, on which a computer program 902 is stored, which program 902, when being executed by a processor, implements the steps described in the foregoing method embodiment for similar image retrieval.
Based on the same inventive concept, embodiments of the present invention provide a computing device 1900, as illustrated with reference to fig. 10, where the computing device 1900 may vary significantly due to different configurations or capabilities, and may include one or more Central Processing Units (CPUs) 1922 (e.g., one or more processors) and memory 1932, one or more storage media 1930 (e.g., one or more mass storage devices) storing applications 1942 or data 1944. Among other things, the memory 1932 and storage medium 1930 can be transient or persistent storage. The program stored in storage medium 1930 may include one or more modules (not shown), each of which may include a sequence of instructions operating on a computer device. Further, the central processor 1922 may be configured to communicate with the storage medium 1930, and execute a series of instruction operations in the storage medium 1930 on the mobile intelligent terminal 1900. The processor, when executing the program, performs the steps as set forth in any of the foregoing method embodiments for similar image retrieval.
Since the computing device described in this embodiment is a computing device used for implementing the method for retrieving similar images in this embodiment, a specific implementation manner of the computing device of this embodiment and various variations thereof can be understood by those skilled in the art based on the method for retrieving similar images described in this embodiment, and therefore, a detailed description of how to implement the method in this embodiment by the computing device is not provided here. The computing device used by a person skilled in the art to implement the method for similar image retrieval in the embodiments of the present application is within the scope of the present application.
One or more technical schemes provided in the embodiment of the invention have at least the following technical effects or advantages:
1. the invention adopts the image retrieval method of the fusion of the two-view-angle characteristics of the depth sparse self-coding operator aiming at the color characteristics and the texture characteristics, thereby avoiding the inaccuracy of image identification by adopting single characteristics in the prior art, obtaining more accurate identification results and effectively improving the identification rate of image retrieval;
2. the method firstly extracts color features and texture features of the image, and then extracts the features by using a CCA (Canonical correlation analysis) family algorithm; then, carrying out sparse representation on the features by using a DSAE (Doherty Sommers architectures Engineers depth sparse self-coding operator), thereby reducing the overall data volume of the image and completing the feature fusion of two visual angles in a parallel or serial mode; and then calculating the correlation of the retrieval images, and selecting the images at the front in the sequencing result as retrieval results. Therefore, the method avoids the operation complexity increased by multi-level iteration in the prior art, solves the technical problems of high iteration overhead and low operation efficiency of massive image retrieval, improves the calculation speed and greatly reduces the occupation of storage space.
The algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose systems may also be used with the teachings herein. The required structure for constructing such a system will be apparent from the description provided above. Moreover, the present invention is not directed to any particular programming language. It is appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any descriptions of specific languages are provided above to disclose the best mode of the invention.
In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: that the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.
Those skilled in the art will appreciate that the modules in the device in the embodiments may be adaptively changed and disposed in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component and furthermore may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.
The various component embodiments of the invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It will be appreciated by those skilled in the art that a microprocessor or Digital Signal Processor (DSP) may be used in practice to implement some or all of the functionality of some or all of the components for a similar image retrieval system according to embodiments of the present invention. The present invention may also be embodied as apparatus or device programs (e.g., computer programs and computer program products) for performing a portion or all of the methods described herein. Such programs implementing the present invention may be stored on computer-readable media or may be in the form of one or more signals. Such a signal may be downloaded from an internet website or provided on a carrier signal or in any other form.
Claims (10)
1. A method for image recognition, comprising:
acquiring an image to be identified and an image data set;
extracting color features and texture features of the image to be recognized to generate color feature vectors and texture feature vectors of the image to be recognized;
performing CCA family correlation analysis calculation on the image data set aiming at the color characteristic vector and the texture characteristic vector of the image to be identified so as to generate a color characteristic matrix and a texture characteristic matrix of the image data set;
respectively carrying out fusion processing on the color characteristic vector and the texture characteristic vector of the image to be identified and the color characteristic matrix and the texture characteristic matrix of the image data set by adopting a parallel or serial connection method to obtain the fusion characteristic vector of the image to be identified and the global characteristic matrix of the image data set;
and calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be identified according to the global characteristic matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining a retrieval result which is most similar to the image to be identified.
2. The method of claim 1, wherein the acquiring the image to be identified and the image dataset comprises:
a user inputs an image to be recognized and an image data set; or
Receiving an image to be identified and an image data set through a preset data transmission interface; or
The user inputs one of the image to be recognized and the image data set, and the other one is received and obtained through a preset data transmission interface.
3. The method of claim 1, wherein the extracting color features and texture features of the image to be recognized to generate a color feature vector and a texture feature vector of the image to be recognized comprises:
and respectively extracting the color features and the texture features of the image to be identified by utilizing a color histogram and a Gabor feature extraction method so as to generate the color feature vector and the texture feature vector of the image to be identified.
4. The method according to claim 1, wherein after performing CCA-family correlation analysis computation on the image dataset with respect to the color feature vector and the texture feature vector of the image to be recognized to generate the color feature matrix and the texture feature matrix of the image dataset, before performing fusion processing on the color feature vector and the texture feature vector of the image to be recognized and the color feature matrix and the texture feature matrix of the image dataset by using a parallel or serial method to obtain a fusion feature vector of the image to be recognized and a global feature matrix of the image dataset, the method further comprises:
and respectively carrying out dimension reduction on the color characteristic matrix and the texture characteristic matrix of the image data set by adopting a standard correlation analysis method to obtain the effective color characteristic matrix and texture characteristic matrix of the image data set after dimension reduction.
5. The method according to any one of claims 1 to 4, wherein after the parallel or serial method is used to perform the fusion processing on the color feature vector and the texture feature vector of the image to be recognized and the color feature matrix and the texture feature matrix of the image data set, respectively, to obtain the fusion feature vector of the image to be recognized and the global feature matrix of the image data set, before the calculating the correlation between the image data set and the fusion feature of the image to be recognized according to the global feature matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining the retrieval result most similar to the image to be recognized, the method further comprises:
and carrying out multilayer feature reconstruction on the global feature matrix of the image data set by using a depth extreme learning machine to obtain the global feature matrix of the image data set in sparse representation.
6. The method according to any one of claims 1 to 4, wherein after performing CCA group correlation analysis calculation on the image dataset with respect to the color feature vector and the texture feature vector of the image to be recognized to generate the color feature matrix and the texture feature matrix of the image dataset, before performing fusion processing on the color feature vector and the texture feature vector of the image to be recognized and the color feature matrix and the texture feature matrix of the image dataset by using a parallel or serial method to obtain the fusion feature vector of the image to be recognized and the global feature matrix of the image dataset, further comprising:
and performing multilayer feature reconstruction on the color feature matrix and the texture matrix of the image data set by using a deep extreme learning machine to obtain the color feature matrix and the texture feature matrix of the image data set which are expressed sparsely.
7. The method according to claim 1, wherein the calculating the correlation of the fused feature vector of the image data set and the image to be recognized according to the global feature matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining the retrieval result most similar to the image to be recognized comprises:
and calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be recognized according to the global characteristic matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining the image arranged at the first position as the recognition result most similar to the image to be recognized to be output.
8. The method according to claim 1, wherein the calculating the correlation of the fused feature vector of the image data set and the image to be recognized according to the global feature matrix of the image data set, performing descending order arrangement on the images in the image data set, and determining the retrieval result most similar to the image to be recognized comprises:
and calculating the correlation degree of the fusion characteristic vector of the image data set and the image to be identified according to the global characteristic matrix of the image data set, responding to the calculation result of the correlation degree and a preset retrieval amount threshold value, arranging the images in the image data set in a descending order according to the correlation degree, and outputting the image in the preset retrieval amount threshold value which is arranged in the front order as a retrieval result.
9. A computing device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the steps of the method of any one of claims 1-8 are implemented when the program is executed by the processor.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110482595.9A CN113313126A (en) | 2021-04-30 | 2021-04-30 | Method, computing device, and computer storage medium for image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110482595.9A CN113313126A (en) | 2021-04-30 | 2021-04-30 | Method, computing device, and computer storage medium for image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113313126A true CN113313126A (en) | 2021-08-27 |
Family
ID=77371339
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110482595.9A Pending CN113313126A (en) | 2021-04-30 | 2021-04-30 | Method, computing device, and computer storage medium for image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113313126A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115086127A (en) * | 2022-05-26 | 2022-09-20 | 杭州电子科技大学 | Improved MPPSK signal demodulation system based on stack DSAE network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105404657A (en) * | 2015-11-04 | 2016-03-16 | 北京工业大学 | CEDD feature and PHOG feature based image retrieval method |
| CN106126581A (en) * | 2016-06-20 | 2016-11-16 | 复旦大学 | Cartographical sketching image search method based on degree of depth study |
| CN106203346A (en) * | 2016-07-13 | 2016-12-07 | 吉林大学 | A kind of road environment image classification method towards the switching of intelligent vehicle driving model |
| CN108009191A (en) * | 2017-09-24 | 2018-05-08 | 贵州师范学院 | A kind of image search method |
| CN108304588A (en) * | 2018-03-07 | 2018-07-20 | 山东师范大学 | A kind of image search method and system based on k neighbours and Fuzzy Pattern Recognition |
| CN110162654A (en) * | 2018-01-30 | 2019-08-23 | 西安邮电大学 | It is a kind of that image retrieval algorithm is surveyed based on fusion feature and showing for search result optimization |
| US20200294234A1 (en) * | 2019-03-13 | 2020-09-17 | Matchlab, Inc. | Methods and systems for automatedly collecting and ranking dermatological images |
| CN111881849A (en) * | 2020-07-30 | 2020-11-03 | Oppo广东移动通信有限公司 | Image scene detection method and device, electronic equipment and storage medium |
-
2021
- 2021-04-30 CN CN202110482595.9A patent/CN113313126A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105404657A (en) * | 2015-11-04 | 2016-03-16 | 北京工业大学 | CEDD feature and PHOG feature based image retrieval method |
| CN106126581A (en) * | 2016-06-20 | 2016-11-16 | 复旦大学 | Cartographical sketching image search method based on degree of depth study |
| CN106203346A (en) * | 2016-07-13 | 2016-12-07 | 吉林大学 | A kind of road environment image classification method towards the switching of intelligent vehicle driving model |
| CN108009191A (en) * | 2017-09-24 | 2018-05-08 | 贵州师范学院 | A kind of image search method |
| CN110162654A (en) * | 2018-01-30 | 2019-08-23 | 西安邮电大学 | It is a kind of that image retrieval algorithm is surveyed based on fusion feature and showing for search result optimization |
| CN108304588A (en) * | 2018-03-07 | 2018-07-20 | 山东师范大学 | A kind of image search method and system based on k neighbours and Fuzzy Pattern Recognition |
| US20200294234A1 (en) * | 2019-03-13 | 2020-09-17 | Matchlab, Inc. | Methods and systems for automatedly collecting and ranking dermatological images |
| CN111881849A (en) * | 2020-07-30 | 2020-11-03 | Oppo广东移动通信有限公司 | Image scene detection method and device, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 张丽: "基于颜色和纹理特征的图像检索技术研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》 * |
| 郑秋梅，等: "基于FCCA的多特征融合的检索方法", 《电子设计工程》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115086127A (en) * | 2022-05-26 | 2022-09-20 | 杭州电子科技大学 | Improved MPPSK signal demodulation system based on stack DSAE network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6893233B2 (en) | Image-based data processing methods, devices, electronics, computer-readable storage media and computer programs | |
| US20230087526A1 (en) | Neural network training method, image classification system, and related device | |
| Abiyev et al. | Sign language translation using deep convolutional neural networks | |
| Yang | Plant leaf recognition by integrating shape and texture features | |
| Ab Jabal et al. | Leaf features extraction and recognition approaches to classify plant | |
| WO2019018063A1 (en) | Fine-grained image recognition | |
| Chang et al. | A novel content based image retrieval system using k-means/knn with feature extraction | |
| WO2014205231A1 (en) | Deep learning framework for generic object detection | |
| KR20140079262A (en) | Image retrieval method, real-time drawing prompting method and devices thereof | |
| Ning et al. | Occluded person re-identification with deep learning: a survey and perspectives | |
| Garcia-Fidalgo et al. | On the use of binary feature descriptors for loop closure detection | |
| Ding et al. | Dynamic perception framework for fine-grained recognition | |
| CN111400528A (en) | Image compression method, device, server and storage medium | |
| CN114647713A (en) | Knowledge graph question-answering method, device and storage medium based on virtual confrontation | |
| Han et al. | L-Net: lightweight and fast object detector-based ShuffleNetV2 | |
| Singh et al. | Geometrically invariant color, shape and texture features for object recognition using multiple kernel learning classification approach | |
| US11281714B2 (en) | Image retrieval | |
| Wang et al. | Multistage model for robust face alignment using deep neural networks | |
| CN113313126A (en) | Method, computing device, and computer storage medium for image recognition | |
| Yang et al. | Robust visual tracking based on deep convolutional neural networks and kernelized correlation filters | |
| Liu et al. | Deep learning-based image target detection and recognition of fractal feature fusion for BIOmetric authentication and monitoring | |
| CN110209895B (en) | Vector retrieval method, device and equipment | |
| Zhu et al. | Scene text relocation with guidance | |
| Adnan et al. | Automated image annotation with novel features based on deep ResNet50-SLT | |
| Liang et al. | Random forest with suppressed leaves for Hough voting |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210827 |