CN100369047C - Image identifying method based on Gabor phase mode - Google Patents
Image identifying method based on Gabor phase mode Download PDFInfo
- Publication number
- CN100369047C CN100369047C CNB2005100680275A CN200510068027A CN100369047C CN 100369047 C CN100369047 C CN 100369047C CN B2005100680275 A CNB2005100680275 A CN B2005100680275A CN 200510068027 A CN200510068027 A CN 200510068027A CN 100369047 C CN100369047 C CN 100369047C
- Authority
- CN
- China
- Prior art keywords
- gabor
- image
- map
- global
- phase
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The present invention discloses an image identifying method based on Gabor phase modes, which comprises the steps: images are selected; images to be compared are performed Gabor transformation, and Gabor characteristic maps of all images are obtained; global Gabor phase modes of each Gabor characteristic pattern in the Gabor characteristic maps are extracted; local Gabor phase modes of each Gabor characteristic pattern in the Gabor characteristic maps are extracted; the global Gabor phase modes and the local Gabor phase modes are counted, and statistical results are connected in series to be high-dimensional characteristic vectors; the high-dimensional characteristic vectors of the images are compared, the similarity among the high-dimensional characteristic vectors is obtained, and the images are identified by using the similarity. The present invention has the advantages of reservation of the structure information of texture images, high identifying precision and good identifying effects, can be used for identifying complex images, can effectively overcome the affect of interference factors of light irradiation, gestures, aging, noise, etc.
Description
Technical Field
The invention relates to the field of image identification, in particular to an image identification method based on a Gabor phase mode.
Background
With the advent of digital images, the need to process the images, an important aspect of image processing is image recognition. Image recognition is to determine whether two or more images are the same image or whether the images are substantially the same through comparison between the images. Image recognition has been widely used and successful in biometric identification, such as fingerprint identification, face identification, iris identification, and the like.
Face recognition is a successful application of image recognition technology, and has received wide attention in the fields of commercial application and research. Specifically, the face recognition is to use one face image as a standard to recognize another face image, and see whether the faces in the two face images are the same face. In the process of face recognition, the face image is inevitably influenced by interference factors such as posture, illumination, aging, noise and the like, and a good face recognition method needs to overcome the interference as much as possible and accurately recognize the face.
The existing face recognition methods can be divided into two types on the whole, one is a face recognition method based on statistical analysis, and the other is a face recognition method based on template matching. A Face Recognition method based on statistical Analysis is widely applied In the field of Face Recognition, and typical Face Recognition methods based on statistical Analysis include principal component Analysis, linear discrimination technology, bayesian (Bayesian) method, and the like (reference [1]: M.Turk and A.Pentland, "Face registration using information faces", in Proc.IEEE Conference on Computer Vision and Pattern registration, 1991, pp.586-591. Reference [2]: P.Belhummer, P.Hespanha, and D.Kriegman, "Eigenfaecs vs. Fishface: registration using class specific knowledge protocol", IEEE Transactions on Pattern Analysis and Machine Analysis, 1997, 19 (7), pp.711-720. Reference [3], B.Moghaddress, C.Nastar, A.Pentland, "A Bayesian location for use In reading, in 3 registration, in 1 registration, 350. Registration, 1996, in protocol, P.I.I.I.I. and P.II.. The above methods based on statistical analysis all have the disadvantage of weak generalization ability, that is, these methods need a large amount of training data to train the recognition model, but usually the available training data is limited, so the distribution of the training data often cannot reflect the distribution of the test data well, and the recognition result is finally affected.
The face recognition method based on the template matching is provided aiming at the defect that the face recognition method based on the statistical analysis is not strong in generalization capability. The face recognition method based on template matching is to encode face images by using a uniform template and then realize face recognition through matching between codes. One specific application example of the face recognition method based on template matching is to realize face recognition by Gabor transformation.
The Gabor transform is a short-time fourier transform with a Gaussian (Gaussian) function as a window function, the basic idea of which is to divide the signal into a number of small time intervals, each time interval being analyzed by the fourier transform in order to determine the frequency at which the time interval exists. The Gabor filter using the Gabor transformation principle can simultaneously retain information of a space domain and a frequency domain, and therefore, the method is applied to a face recognition method based on template matching. The specific implementation of Gabor transformation is to perform convolution operation on Gabor wavelets and image gray level images to obtain a Gabor characteristic spectrum. The Gabor wavelet can be represented by equation (1):
Ψ μ，，v (z)＝(‖K μ，v ‖/σ 2 )exp(‖K μ，v ‖ 2 ‖z‖ 2 /σ 2 )(exp(i.K μ，v z)-exp(-σ 2 2)) (1) where |, represents the modulo operation, z = (x, y) represents the position of the pixel in the space domain, σ is the standard deviation of a gaussian function along the x-and y-axes, K μ，v ＝(k v cos(Φ μ )，k v sin(Φ μ ))，k v ＝2 -(v+2)π/2 ，Φ μ μ π/8, μ represents the scale, v represents the Gabor wavelet direction. The scale referred to herein is a quantity used to represent frequency, in a sense different from the usual scale. The values of the dimension μ and the direction v are variable in specific applications, typically μ =0,1, ·,4,v =0,1, ·,7 (detailed information about wavelet calculation formulas can be found in the reference [4 ] see the literature for details on wavelet calculation formulas]: chengjunLiu and hurrywechsler, gaborfeaturebaseedClassification Using the enhancement Fisher LineardesinsentModelfor faceRecognization. IEEETransImageprocessing vol.11 No.4, (2002) 467-476). Let I (z) represent the gray distribution of the face image, which can be obtained by graying the image. Image I (z) and Gabor wavelet Ψ μ，v (z) the convolution formula is
G μ，v (z)＝I(z)*Ψ μ，v (z) (2)
Here denotes a convolution operation.
After the Gabor feature map is obtained by Gabor transformation, in the prior art, a Face Recognition method is to perform Face Recognition by using Gabor magnitude information (the detailed information about Face Recognition by using Gabor magnitude information can be found in reference [5]: w.y.zhao, r.chellappa, p.j.phillips and a.rosenfeld, "Face Recognition: a light characterization Survey", ACM Computing Survey 2003, pp.399-458.). Another direction of thinking for face recognition using Gabor transform is to use Gabor phase information, which includes real part information and imaginary part information. Compared with Gabor amplitude information, gabor phase information contains more information, and the accuracy of face recognition is improved.
In the prior art, gabor phase information is already applied to iris recognition and obtains good recognition effect. In the method for iris recognition using Gabor phase information, re (G) is used μ，v (z)) and Im (G) μ，v (z)) respectively represent Gabor profiles G μ，v The real and imaginary parts of (z). By P μ，v Re (z) and P μ，v Im (z) represents the real and imaginary parts of the Gabor phase information, respectively, which are the result of the Gabor profile after quantization. The principle of Gabor characteristic spectrum quantization is as follows: if the real part of the Gabor feature map is greater than 0, the real part of the phase information is 0, and if the real part of the Gabor feature map is less than or equal to 0, the real part of the phase information is 1; the same is true for the imaginary part. The quantization calculation formula is shown in formula (3):
according to the above formula, the quantized result can be concatenated into a binary character string, the character string is used as the feature to be extracted finally, and then the face recognition is performed by using the hamming distance as the similarity formula in the recognition process (detailed information of the iris recognition method using Gabor phase information can be found in reference [6]: J.G. Daugman, 'High reliability visual recognition of individuals by a time of the statistical index', IEEE Transaction on Pattern Analysis and Machine Analysis, 1993, vol.15, pp.1148-1161). The method utilizes Gabor phase information to realize iris recognition, but the method has the problem of too simple mode and cannot be used for recognizing more complex images (such as human faces), thereby limiting the application range of the method.
Disclosure of Invention
The invention aims to provide an image identification method based on a Gabor phase mode, which realizes the identification of complex images.
In order to achieve the above object, the present invention provides an image recognition method based on a Gabor phase mode, including:
step 1), an image selection step, namely selecting images to be compared;
step 2) Gabor transformation is carried out on the images to be compared, each image obtains a respective Gabor characteristic map spectrum, and the Gabor characteristic maps in the Gabor characteristic maps are divided into Gabor characteristic maps based on a real part and Gabor characteristic maps based on an imaginary part;
step 3) extracting a global Gabor phase mode from each Gabor characteristic map obtained in the step 2), wherein the global Gabor phase mode is divided into a global Gabor phase mode based on a real part Gabor characteristic map and a global Gabor phase mode based on an imaginary part Gabor characteristic map;
step 4) extracting a local Gabor phase mode from each Gabor characteristic map obtained in the step 2), wherein the local Gabor phase mode is divided into a local Gabor phase mode based on a real part Gabor characteristic map and a local Gabor phase mode based on an imaginary part Gabor characteristic map;
step 5) carrying out statistics on the global Gabor phase mode obtained in the step 3) and the local Gabor phase mode obtained in the step 4), and connecting the statistical result in series into high-dimensional feature vectors, wherein each image has a corresponding high-dimensional feature vector;
and 6) comparing the high-dimensional feature vectors of the images to obtain the similarity among the high-dimensional feature vectors, and identifying the images by using the similarity.
In the above technical solution, in the step 1), when the image is selected, the image is segmented to segment the core region of the image, and only the core region of the image is Gabor transformed during subsequent Gabor transformation.
In the above technical solution, in the step 2), the Gabor transform is a convolution operation performed on a Gabor wavelet and an image gray scale map, during the Gabor transform, the dimension and direction of the Gabor wavelet have a plurality of values, and different Gabor feature maps are obtained from different values, so as to obtain the Gabor feature map.
In the above technical solution, in the step 3), extracting the global Gabor phase pattern includes:
a1 In the Gabor characteristic map, selecting a point with the same position on each Gabor characteristic map with the same scale and different directions, and quantizing the Gabor phase information of the point to obtain a quantization result of 1-bit binary number in each Gabor characteristic map;
a2 Quantizing Gabor phase information of points at other positions on each Gabor characteristic diagram with the same scale and different directions to obtain a plurality of matrixes consisting of binary numbers, wherein the number of the matrixes is the same as that of the Gabor characteristic diagrams with the same scale and different directions;
a3 B) performing series connection on binary numbers of the same position of each matrix obtained in the step a 2) to obtain a global Gabor phase pattern, wherein if the Gabor characteristic diagram based on the quantization processing is a real part Gabor characteristic diagram, the global Gabor phase pattern is the global Gabor phase pattern based on the real part Gabor characteristic diagram, and if the Gabor characteristic diagram based on the quantization processing is an imaginary part Gabor characteristic diagram, the global Gabor phase pattern is the global Gabor phase pattern based on the imaginary part Gabor characteristic diagram.
In the above technical solution, in the step 4), extracting the local Gabor phase pattern includes:
b1 ) carrying out quantization processing on each Gabor characteristic map in the Gabor characteristic maps to obtain a quantization result of a 1-bit binary number;
b2 Selecting a point in a Gabor characteristic map, performing exclusive OR operation on the quantization result of the point in the step b 1) and the quantization results of eight points around the point in the step b 1), sequentially connecting the eight exclusive OR operation results in a binary number string with 8 bits, wherein the binary number string with 8 bits represents the selected point, and performing the same operation on other points in the Gabor characteristic map to obtain a local Gabor phase pattern of the Gabor characteristic map;
b3 Other Gabor feature maps in the Gabor feature map are selected, and the step b 2) is repeated until all the Gabor feature maps obtain respective local Gabor phase modes.
In the above technical solution, in the step 5), the Gabor phase mode is divided into a plurality of mutually disjoint blocks before statistics of the Gabor phase mode; the blocking method for each Gabor phase mode in a Gabor feature map is the same or different, but the blocking method must be the same between corresponding Gabor phase modes in different Gabor feature maps.
In the above technical solution, in the step 5), the Gabor phase mode is counted by using an extracted histogram method, the counted histograms are connected in series to obtain a high-dimensional histogram, and histogram intersection operations are performed on different high-dimensional histograms to implement comparison between the high-dimensional histograms.
After the Gabor phase mode is blocked, different weights are given to all the blocks before the Gabor phase mode is counted, the weight of the block where the region with obvious influence on the identification effect is located is larger, the weights of the corresponding blocks of the compared image are the same, and in the step 6), the comparison results of the high-dimensional feature vectors are respectively multiplied by the weights to obtain the similarity among the high-dimensional feature vectors.
The invention has the advantages that:
1. the image identification method based on the Gabor phase mode can keep the structure information of the texture image, can be used for identifying complex images, and is high in identification precision.
2. The image identification method based on the Gabor phase mode can effectively overcome the influence of interference factors such as illumination, posture, aging and noise, and has good identification effect.
Drawings
FIG. 1 is the result of the transformation of an image in the Gabor transform domain;
FIG. 2 is a local Gabor phase pattern based on real information;
FIG. 3 is a global Gabor phase pattern based on real information;
FIG. 4 is a schematic diagram of calculating a local Gabor phase pattern;
FIG. 5 is a comparison chart of the recognition effect of the present invention on the expression change test set when used for face recognition and the existing face recognition method;
FIG. 6 is a comparison graph of the recognition effect of the present invention on the illumination change test set when used for face recognition and the existing face recognition method;
FIG. 7 is a comparison graph of the recognition effect of the present invention on the time-varying test set (duplicate-I) when applied to face recognition;
FIG. 8 is a comparison graph of the recognition effect of the present invention on the time-varying test set (duplicate-II) when applied to face recognition;
FIG. 9 is a flow chart of the method of the present invention.
Detailed Description
The Gabor phase mode-based image recognition method according to the present invention is described in detail below with reference to the accompanying drawings and the detailed description thereof.
Taking the most common face recognition in image recognition as an example, as shown in fig. 9, the method of the present invention includes:
step 10: two images are selected, one image is a standard image for comparison, and the other image is an image to be identified. In the embodiment of face recognition, the two selected images respectively include face regions. The standard image is denoted as image 1 and the image to be recognized is denoted as image 2.
Step 20: and respectively performing segmentation operation on the two selected images to segment the core areas of the two images, so that the influence of background parts irrelevant to the core areas in the images on image identification is avoided, and whether the core areas of the images are consistent or not is mainly identified during image identification. In the embodiment of face recognition, when the image 1 and the image 2 are divided, the face can be divided according to the positions of the eyes of the face in the image according to the designated size, so as to obtain the core regions of the face in the image 1 and the image 2, and the core regions of the face in the image 1 and the image 2 are respectively referred to as the face region 1 and the face region 2.
Step 30: and respectively carrying out Gabor transformation on the core regions divided by the two images to obtain Gabor characteristic maps of the respective images. In the embodiment of face recognition, the face region 1 and the face region 2 are respectively subjected to Gabor transformation to obtain a Gabor feature map 1 and a Gabor feature map 2. The specific operations of Gabor transformation for the face region 1 and the face region 2 are substantially the same, and similar transformation results exist, and the Gabor transformation will be briefly described by taking the face region 1 as an example. The Gabor transformation process is as shown in formula (1) and formula (2) in the background art, and a Gabor wavelet is convolved with a gray level image of a human face region 1 to obtain a Gabor feature map 1. In the expression formula (1) of the Gabor wavelet, because the dimension mu and the direction v are variable, if a plurality of values are taken for mu and v, a plurality of Gabor wavelets can be obtained, and a plurality of Gabor feature maps can be obtained. As shown in fig. 1, (a) in the figure represents an original face image, in one embodiment, μ takes five values of 0,1, 2, 3, and 4, v takes eight values of 0,1, 2, 3, 4, 5, 6, and 7, when calculating a Gabor wavelet, the scale is firstly kept unchanged, different values are taken for directions, that is, when μ takes 0, the direction v takes 0,1, 2, 3, 4, 5, 6, and 7 respectively to obtain different Gabor wavelets, the Gabor wavelet and the image gray scale map are convolved to obtain different Gabor feature maps, when μ takes 1, 2, 3, and 4, the direction v takes different values of 0,1, 2, 3, 4, 5, 6, and 7 respectively to obtain different Gabor feature maps, and all the Gabor feature maps are Gabor feature maps. The Gabor feature map can be further divided into a real Gabor feature map and an imaginary Gabor feature map, and (b) in fig. 1 is the real part of the Gabor feature map, which is composed of the real Gabor feature map. The Gabor feature map 1 includes 40 real Gabor feature maps and 40 imaginary Gabor feature maps. The Gabor transform is a mature prior art and can be readily implemented by one of ordinary skill in the art using the formulas and related references provided herein.
Step 40: and respectively extracting global Gabor phase modes from Gabor characteristic maps of the two images, wherein the global Gabor phase modes embody the change rules of the Gabor characteristic maps in different directions under a certain scale. In the embodiment of face recognition, the steps of extracting the global Gabor phase pattern from the Gabor feature map 1 and the Gabor feature map 2 are substantially the same, and therefore, the process of extracting the global Gabor phase pattern will be described by taking the operation on the Gabor feature map 1 as an example.
Step 41: selecting Gabor characteristic maps with the same scale and different directions from the Gabor characteristic map 1, and selecting a point Z with the same position on the characteristic maps 0 . For the selected point Z 0 And quantizing the Gabor phase information to obtain a quantization result. In a specific embodiment, the value of μ is first set to 0 and kept constant, then v is set to a different value, and the quantization result P is obtained according to the quantization calculation formula (3) μ，0 Re (Z 0 )、P μ，1 Re (Z 0 )…… P μ，k Re (Z 0 ) (μ =0, k = 7). The resulting quantization result is a binary number of 0 or 1, i.e., gabor phase information of the point.
Step 42: and performing the same quantization processing on other points of the Gabor characteristic diagram with the same scale and different directions on the Gabor characteristic diagram 1 to obtain a quantization result similar to that in the step 41. And after all points of the Gabor characteristic diagrams with the same scale and different directions on the Gabor characteristic diagram 1 are quantized, obtaining a plurality of matrixes with the same size as the Gabor characteristic diagrams, wherein one point on the Gabor characteristic diagrams corresponds to a binary number at a corresponding position in the matrixes. The number of the matrices is related to the number of directions that can be taken when the dimension is unchanged in the Gabor feature map 1, and in a specific embodiment, the direction v in the real part Gabor feature map can take 8 different values, so that 8 matrices can be obtained. In each matrix, the direction of the points represented by each number is the same, and the directions of the points represented by the numbers of different matrices are different. Likewise, the imaginary Gabor feature map may also result in 8 matrices.
Step 43: and (4) carrying out serial connection on Gabor phase information in different directions of the same scale to obtain a global Gabor phase mode of a real part Gabor characteristic diagram and a global Gabor phase mode of an imaginary part Gabor characteristic diagram. With GGPP μ Re (Z 0 ) Global Gabor phase pattern representing real Gabor feature map, using GGPP μ Im (Z 0 ) A global Gabor phase pattern representing a dashed Gabor profile. The extraction of the global Gabor phase pattern is shown in equation (4):
in one embodiment, 1-bit binary numbers of corresponding positions of 8 matrixes of the real part Gabor characteristic diagram are concatenated together to obtain a new matrix, each element in the matrix consists of 8-bit binary character strings, and the concatenation sequence of the 1-bit binary numbers is related to the direction. The same process is also performed on the 8 matrices of the real Gabor feature map.
Global Gabor phase pattern GGPP of real Gabor feature map μ Re (Z 0 ) Global Gabor phase pattern GGPP with imaginary Gabor feature maps μ Im (Z 0 ) It can also be expressed in decimal system, and its calculation formula is shown in formula (5):
after the phase information in different directions of the same scale is concatenated, the point with a certain scale on the Gabor characteristic map can be represented by 0-255 decimal system.
Step 44: and changing the scale value of the Gabor characteristic diagram, repeating the operations of quantization, extraction and concatenation on the characteristic diagrams in different directions on the scale value to obtain the global Gabor phase mode of the Gabor characteristic diagram corresponding to the scale until the global Gabor phase modes of all scales in the Gabor characteristic diagram 1 are obtained. In one embodiment, the value of the scale μ is changed, the values of μ are taken as 1, 2, 3, 4, respectively, and steps 41, 42, and 43 are repeated to obtain the global Gabor phase pattern. As shown in fig. 3, the global Gabor phase pattern of the real Gabor feature map, 5 of fig. 3 have different scales.
After the extraction of the global Gabor phase mode is completed on the Gabor characteristic diagram 1, 10 matrixes with the same size as the Gabor characteristic diagram are finally obtained, and in the 10 matrixes, 5 matrixes relate to a real part Gabor characteristic diagram, and the other 5 matrixes relate to an imaginary part Gabor characteristic diagram. In the matrix of the real Gabor feature map (or imaginary Gabor feature map), different matrices represent different scales. The extraction of the global Gabor phase pattern for Gabor feature map 2 is similar to Gabor feature map 1.
Step 50: and respectively extracting local Gabor phase modes from Gabor characteristic maps of the two images, wherein the local Gabor phase modes embody the relation between points in the images and surrounding points. In the face recognition embodiment, the steps of extracting the local Gabor phase modes from the Gabor feature map 1 and the Gabor feature map 2 are substantially the same, and therefore, the local Gabor phase mode extraction process will be described by taking the operation on the Gabor feature map 1 as an example.
Step 51: quantizing each real part Gabor characteristic map in the Gabor characteristic map 1 according to a quantization calculation formula (3) to obtain a quantization result P μ，v Re (Z) (μ =1, 2, 3, 4 v =0,1, 2, 3, 4, 5, 6, 7), wherein Z represents any point in the figure; quantizing each imaginary part Gabor characteristic diagram in the Gabor characteristic diagram 1 according to a quantization calculation formula (3) to obtain a quantization result P μ，v Im (Z) (μ =1, 2, 3, 4 v =0,1, 2, 3, 4, 5, 6, 7), where Z represents any point in the diagram.
Step 52: selecting a certain Gabor characteristic map in the Gabor characteristic map 1, and selecting a point Z in the map no matter whether the map is a real part Gabor characteristic map or an imaginary part Gabor characteristic map 0 As shown in FIG. 4, points around the point are respectively marked as Z 1 To Z 8 Selecting a point Z 0 And carrying out XOR operation with surrounding points, wherein the result of the XOR operation is 8 1-bit binary numbers, and the 8 1-bit binary numbers are arranged according to the surrounding points Z 1 To Z 8 The 8-bit binary string is the local Gabor phase information of the point, and the local Gabor phase information can also be converted into decimal numbers to be represented. With LGPP μ，v Re (Z 0 ) Indicating point Z 0 Local Gabor phase information based on real part information, with LGPP μ，v Im (Z 0 ) Indicating point Z 0 Based on the local Gabor phase information of the imaginary part information, the specific formula of the solution is shown as formula (6), where XOR represents exclusive or operation:
the selected point Z can be obtained by the above operation 0 According to the local Gabor phase information, local Gabor phase information of other points in a certain Gabor characteristic diagram can be obtained according to the same method, the local Gabor phase information of all the points which can be obtained is expressed by a matrix, and the obtained matrix is the local Gabor phase mode. The local Gabor phase information for points at the upper, lower, left and right boundaries in the Gabor profile need not be found.
Step 53: and selecting other Gabor feature maps in the Gabor feature map 1, and performing the same operation as the step 52 on the feature maps until all Gabor feature maps in the Gabor feature map 1 are subjected to similar operation. As shown in fig. 2, a local Gabor phase pattern of the real Gabor characteristic diagram. The dimensions and orientations of the various figures in fig. 2 are different.
By operating on the extracted local Gabor phase mode of the Gabor feature map 1, 40 local Gabor phase modes based on the real part Gabor feature map and 40 local Gabor phase modes based on the imaginary part Gabor feature map can be obtained. The same is true for the operation of Gabor profile 2.
Step 60: and (4) performing partitioning operation on all global Gabor phase modes and local Gabor phase modes of the two images, wherein the partitioned blocks are not intersected with each other. In an embodiment of face recognition, each image may obtain 10 global Gabor phase patterns and 80 local Gabor phase patterns, the phase patterns are blocked, and the methods for blocking between the global Gabor phase patterns and between the local Gabor phase patterns may be the same or different, that is, the number of blocks may be different, and the size of each block may also be different, but the blocking method between the phase patterns corresponding to the two images for recognition must be the same. The corresponding phase modes mentioned here mean that the phase modes to be referred to must be based on the real Gabor characteristic diagram or the imaginary Gabor characteristic diagram, and the scales and directions of the Gabor characteristic diagrams represented by the phase modes to be referred to are the same.
Step 70: and counting the phase information on each block, and connecting the counting results of all the blocks into a high-dimensional feature vector. In one embodiment of face recognition, statistical phase information may be implemented by extracting a histogram. The operation process of extracting the histogram is to take a global Gabor phase pattern based on real part information as an example, the phase pattern contains global Gabor phase information of each point on the image at a certain scale, if the phase information is represented by decimal system and ranges from 0 to 255, the global Gabor phase pattern is divided into 10 blocks, the occurrence times of 256 numbers of 0 to 255 are counted in each block, and the histogram of a block is that the occurrence times of the 256 numbers in the block are listed on the graph respectively according to the sequence. A phase pattern is divided into 10 blocks to list 10 histograms, and then the 10 histograms are concatenated side by side to obtain the histogram of the phase pattern. The operation of extracting histograms of other Gabor phase modes is similar to the operation, and finally, the histograms of all Gabor phase modes (including a global Gabor phase mode and a local Gabor phase mode) of one image are connected in series to obtain a high-dimensional histogram so as to realize the coding of the face image. In image recognition, the concatenation order of the histograms in the two images should be the same. The Gabor characteristic map 1 obtains a high-dimensional histogram 1, and the Gabor characteristic map 2 obtains a high-dimensional histogram 2. Extraction of histograms and concatenation of high-dimensional histograms are well known in the art and can be readily implemented by those skilled in the art.
Step 80: after the two images obtain respective high-dimensional feature vectors, the high-dimensional feature vectors are compared to obtain a similarity, the images are identified according to the similarity, and the higher the value of the similarity is, the more similar the two images are. In the embodiment of face recognition, the high-dimensional histogram 1 and the high-dimensional histogram 2 are compared, and face recognition is performed by adopting a histogram matching method to obtain the similarity between the image 1 and the image 2. The histogram matching method can be implemented by using a histogram intersection formula, as shown in formula (7):
wherein H 1 And H 2 Representing a high-dimensional histogram 1 and a high-dimensional histogram 2, H 1 i And H 2 i A comparison term representing the high-dimensional histogram 1 and the high-dimensional histogram 2, which is a histogram of each block. And taking the smaller value of the comparison items of the two high-dimensional histograms, and finally adding all the values to obtain the similarity of the image 1 and the image 2. Histogram matching is well established prior art and will not be described in detail herein.
After the similarity between the image 1 and the image 2 is obtained, the human face can be recognized by utilizing the similarity, a threshold value can be preset in the application of human face recognition, if the similarity between the image 1 and the image 2 is higher than the threshold value, the human faces in the image 1 and the image 2 are the human faces of the same person, and the two human faces can be different in light, age and posture.
The steps describe the specific operation steps of image recognition between two images, in practical application, the method of the invention can also be used for recognizing a plurality of images, the specific process of image recognition is the same as that of image recognition of two images, and in the process of recognizing a plurality of images, the image with the maximum similarity is selected as the final recognition result.
During image recognition, the influence of certain areas in the image on the recognition effect is obvious, and in order to improve the image recognition effect, a preferred scheme of the method is as follows: after the blocking operation of the Gabor phase mode of the image is performed in step 60, different weights w are given to the respective blocks, wherein the weight of the block in which the region having a significant influence on the recognition effect is located is larger, and the weight of the corresponding block of the compared image should be the same. After the blocks are weighted, in the process of calculating the similarity in step 80, the comparison results of the high-dimensional feature vectors are multiplied by the weights w, respectively. Taking face recognition as an example, when histogram matching is realized by adopting a histogram intersection method, a histogram intersection formula is shown as a formula (8):
the use of weights improves the recognition rate of the image.
The method of the invention can effectively solve the influence of disturbing factors such as gesture, illumination, aging, noise and the like in the process of face recognition, and obtain good recognition effect. The method of the invention is tested on a FERET face database, and is compared with the recognition effects of a plurality of existing famous face recognition systems UMD _97, USC _97 and MIT _96, the recognition effects are shown in figures 5, 6, 7 and 8, wherein figure 5 is an expression change test set, figure 6 is an illumination change test set, figures 7 and 8 are time change test sets, GPP in each figure represents the method of the invention, and the method of the invention is obviously improved on each test set compared with the existing face recognition system. Taking the illumination variation test set of fig. 6 as an example, in the test set, the average recognition rate of the method (GPP) of the present invention is close to 100%, the recognition rate of USC _97 is between 80% and 95%, the recognition rate of UMD _97 is between 60% and 90%, and the recognition rate of MIT _96 is only between 30% and 75%.
The method is not only suitable for the field of face recognition, but also suitable for other fields of image recognition, and can obtain good recognition effect.
Claims (8)
1. An image identification method based on a Gabor phase mode comprises the following steps:
step 1), selecting an image, namely selecting the image to be compared;
step 2), gabor transformation is carried out on the images to be compared, each image obtains a respective Gabor characteristic map spectrum, and the Gabor characteristic maps in the Gabor characteristic maps are divided into Gabor characteristic maps based on a real part and Gabor characteristic maps based on an imaginary part;
step 3), extracting a global Gabor phase mode from each Gabor characteristic map obtained in the step 2), wherein the global Gabor phase mode is divided into a global Gabor phase mode based on a real part Gabor characteristic map and a global Gabor phase mode based on an imaginary part Gabor characteristic map;
step 4), extracting a local Gabor phase mode from each Gabor characteristic map in the Gabor characteristic map obtained in the step 2), wherein the local Gabor phase mode is divided into a local Gabor phase mode based on a real part Gabor characteristic map and a local Gabor phase mode based on an imaginary part Gabor characteristic map;
step 5), carrying out statistics on the global Gabor phase mode obtained in the step 3) and the local Gabor phase mode obtained in the step 4), and connecting the statistical results in series into high-dimensional feature vectors, wherein each image has a corresponding high-dimensional feature vector;
and 6) comparing the high-dimensional feature vectors of the images to obtain the similarity among the high-dimensional feature vectors, and identifying the images by using the similarity.
2. The method according to claim 1, wherein in the step 1), the image is segmented to segment a core region of the image when the image is selected, and only the core region of the image is Gabor transformed during subsequent Gabor transformation.
3. The Gabor phase pattern-based image recognition method according to claim 1, wherein in the step 2), the Gabor transform is a convolution operation of a Gabor wavelet and an image gray map, during the Gabor transform, the dimension and the direction of the Gabor wavelet have a plurality of values, and different Gabor feature maps are obtained from different values, so as to obtain the Gabor feature map.
4. The method according to claim 1, wherein the extracting the global Gabor phase pattern in step 3) comprises:
a1 In the Gabor characteristic map, selecting a point with the same position on each Gabor characteristic map with the same scale and different directions, and quantizing the Gabor phase information of the point to obtain a quantization result of 1-bit binary number in each Gabor characteristic map;
a2 Quantizing Gabor phase information of points at other positions on each Gabor characteristic diagram with the same scale and different directions to obtain a plurality of matrixes consisting of binary numbers, wherein the number of the matrixes is the same as that of the Gabor characteristic diagrams with the same scale and different directions;
a3 Binary numbers of the same position of each matrix obtained in the step a 2) are concatenated to obtain a global Gabor phase pattern, if the Gabor feature map based on the quantization processing is a real part Gabor feature map, the global Gabor phase pattern is a global Gabor phase pattern based on the real part Gabor feature map, and if the Gabor feature map based on the quantization processing is an imaginary part Gabor feature map, the global Gabor phase pattern is a global Gabor phase pattern based on the imaginary part Gabor feature map.
5. The method according to claim 1, wherein the extracting local Gabor phase pattern in step 4) comprises:
b1 Carrying out quantization processing on each Gabor characteristic map in the Gabor characteristic map to obtain a quantization result of 1-bit binary number;
b2 For a certain Gabor characteristic map in the Gabor characteristic map, selecting a point in the map, performing exclusive OR operation on the quantization result of the point obtained in the step b 1) and the quantization results of eight points around the point obtained in the step b 1), and sequentially connecting the eight exclusive OR operation results in series to form an 8-bit binary number string, wherein the 8-bit binary number string represents the selected point, and performing the same operation on other points in the Gabor characteristic map to obtain a local Gabor phase pattern of the Gabor characteristic map;
b3 Other Gabor feature maps in the Gabor feature map are selected, and the step b 2) is repeated until all the Gabor feature maps obtain respective local Gabor phase modes.
6. The method according to claim 1, wherein in the step 5), the Gabor phase pattern is divided into a plurality of mutually non-intersecting blocks before statistics of the Gabor phase pattern; the blocking methods of the respective Gabor phase modes in the Gabor feature maps are the same or different, but the blocking methods between corresponding Gabor phase modes in different Gabor feature maps must be the same.
7. The Gabor phase mode-based image recognition method of claim 1, wherein in the step 5), the Gabor phase mode is counted by using an extracted histogram method, the counted histograms are concatenated to obtain a high-dimensional histogram, and histogram inter-operation is performed on different high-dimensional histograms to realize comparison between the high-dimensional histograms.
8. The Gabor phase mode-based image recognition method of claim 6, wherein after the Gabor phase mode is blocked, different weights are given to the blocks before the Gabor phase mode is counted, the weight of the block in which the region with obvious influence on the recognition effect is located is larger, the weights of the corresponding blocks of the compared image are the same, and in the step 6), the comparison results of the high-dimensional feature vectors are multiplied by the weights respectively to obtain the similarity among the high-dimensional feature vectors.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100680275A CN100369047C (en) | 2005-04-30 | 2005-04-30 | Image identifying method based on Gabor phase mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100680275A CN100369047C (en) | 2005-04-30 | 2005-04-30 | Image identifying method based on Gabor phase mode |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1858773A CN1858773A (en) | 2006-11-08 |
| CN100369047C true CN100369047C (en) | 2008-02-13 |
Family
ID=37297681
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2005100680275A Active CN100369047C (en) | 2005-04-30 | 2005-04-30 | Image identifying method based on Gabor phase mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100369047C (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101964113A (en) * | 2010-10-02 | 2011-02-02 | 上海交通大学 | Method for detecting moving target in illuminance abrupt variation scene |
| CN102156872B (en) * | 2010-12-29 | 2013-03-13 | 深圳大学 | Multispectral-data-based object identification method and device |
| CN103489004A (en) * | 2013-09-30 | 2014-01-01 | 华南理工大学 | Method for achieving large category image identification of deep study network |
| CN104504679A (en) * | 2014-11-19 | 2015-04-08 | 国家电网公司 | On-line analysis method of infrared defect on the basis of typical defect map |
| WO2016110125A1 (en) * | 2015-01-09 | 2016-07-14 | 北京大学 | Hash method for high dimension vector, and vector quantization method and device |
| US10460153B2 (en) * | 2016-11-15 | 2019-10-29 | Futurewei Technologies, Inc. | Automatic identity detection |
| CN110532851B (en) * | 2019-07-04 | 2022-04-15 | 珠海格力电器股份有限公司 | Finger vein identification method and device, computer equipment and storage medium |
| CN112213782B (en) * | 2020-09-29 | 2022-03-04 | 中国石油大学(北京) | Processing method and device for sub-phase seismic data and server |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010031103A1 (en) * | 1999-12-03 | 2001-10-18 | Mun-Churl Kim | Texture description method and texture-based image retrieval method using gabor filter in frequency domain |
| CN1459761A (en) * | 2002-05-24 | 2003-12-03 | 清华大学 | Character identification technique based on Gabor filter set |
| US20040175057A1 (en) * | 2003-03-04 | 2004-09-09 | Thomas Tsao | Affine transformation analysis system and method for image matching |
-
2005
- 2005-04-30 CN CNB2005100680275A patent/CN100369047C/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010031103A1 (en) * | 1999-12-03 | 2001-10-18 | Mun-Churl Kim | Texture description method and texture-based image retrieval method using gabor filter in frequency domain |
| CN1459761A (en) * | 2002-05-24 | 2003-12-03 | 清华大学 | Character identification technique based on Gabor filter set |
| US20040175057A1 (en) * | 2003-03-04 | 2004-09-09 | Thomas Tsao | Affine transformation analysis system and method for image matching |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1858773A (en) | 2006-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100369047C (en) | Image identifying method based on Gabor phase mode | |
| Nogueira et al. | Evaluating software-based fingerprint liveness detection using convolutional networks and local binary patterns | |
| Bae et al. | Iris feature extraction using independent component analysis | |
| CN106529447B (en) | Method for identifying face of thumbnail | |
| US7302087B2 (en) | Daubechies wavelet transform of iris image data for use with iris recognition system | |
| Pflug et al. | A comparative study on texture and surface descriptors for ear biometrics | |
| CN109145745B (en) | Face recognition method under shielding condition | |
| CN103839042B (en) | Face identification method and face identification system | |
| Wang et al. | Weber local descriptors with variable curvature gabor filter for finger vein recognition | |
| Mohamed et al. | An improved LBP algorithm for avatar face recognition | |
| JP7130905B2 (en) | Fast and Robust Dermatoglyphic Mark Minutia Extraction Using Feedforward Convolutional Neural Networks | |
| CN108509843A (en) | A kind of face identification method of the Huber constraint sparse codings based on weighting | |
| Dey et al. | Improved feature processing for iris biometric authentication system | |
| Chirchi et al. | Feature extraction and pupil detection algorithm used for iris biometric authentication system | |
| CN111275070B (en) | Signature verification method and device based on local feature matching | |
| CN105631441A (en) | Human face recognition method | |
| Xu et al. | A novel method for iris feature extraction based on intersecting cortical model network | |
| El-Naggar et al. | Which dataset is this iris image from? | |
| US7436999B2 (en) | Data analysis device and data recognition device | |
| CN111931757A (en) | Finger vein quick sorting method and device based on MDLBP block histogram and PCA dimension reduction | |
| CN111488811A (en) | Face recognition method and device, terminal equipment and computer readable medium | |
| Kekre et al. | Performance comparison of DCT and VQ based techniques for iris recognition | |
| CN106326827A (en) | Palm vein recognition system | |
| Rathgeb et al. | Improvement of iris recognition based on iris-code bit-error pattern analysis | |
| Al-Rawi et al. | Feature Extraction of Human Facail Expressions Using Haar Wavelet and Neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |