CN104915626A - Face identification method and apparatus - Google Patents
Face identification method and apparatus Download PDFInfo
- Publication number
- CN104915626A CN104915626A CN201410088084.9A CN201410088084A CN104915626A CN 104915626 A CN104915626 A CN 104915626A CN 201410088084 A CN201410088084 A CN 201410088084A CN 104915626 A CN104915626 A CN 104915626A
- Authority
- CN
- China
- Prior art keywords
- eigenface
- eigenwert
- facial image
- holotype
- negative mode
- 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
Landscapes
- Image Analysis (AREA)
- Studio Devices (AREA)
Abstract
The invention discloses a face identification method and apparatus, for improving the accuracy of face identification. The method comprises the following steps: determining a characteristic value of an LTP self-adaptive threshold of each pixel point in an image to be identified, wherein the self-adaptive threshold corresponding to each pixel point is determined by a gray scale difference between the pixel point and each pixel point in a neighborhood; according to the characteristic value of the LTP self-adaptive threshold, determining a positive-mode characteristic value and a negative-mode characteristic value; determining a positive-mode characteristic face and a negative-mode characteristic face, wherein the positive-mode characteristic face is composed of the positive-mode characteristic value of each pixel point, and the negative-mode characteristic face is composed of the negative-mode characteristic value of each pixel point; calculating a characteristic value histogram of each characteristic face and an information entropy weight of the characteristic face of each layer, and by use of the information entropy weights, performing weight cascading on the characteristic value histograms of the positive-mode and negative-mode characteristic faces to obtain an enhanced histogram; and calculating a chi-square distance between the enhanced histogram of the face image to be identified and the enhanced histogram of each face image with known identity, and according to the chi-square distance, determining an identification result.
Description
Technical field
The present invention relates to mode identification technology, be specifically related to a kind of face identification method and device.
Background technology
Recognition of face is a kind of biological identification technology carrying out identification based on the face feature information of people, and achieve application in some fields in recent years, such as recognition of face can be applied to gate control system, attendance checking system, smart mobile phone etc.
In face recognition technology, mainly contain two steps: from facial image to be identified, extract proper vector; The proper vector of image in proper vector and face database is carried out contrast and obtains recognition result.Wherein, first step directly affects the accuracy of face recognition result.In the prior art, face recognition algorithms is a lot, such as based on LTP(Local ternary Patterns, local three binarization modes) operator carries out the face recognition algorithms of feature extraction, LTP operator is to LBP(Local Binary Patterns, local binary patterns) improvement of operator is the operator of Description Image textural characteristics in a kind of tonal range.But LTP operator adopts self-defined fixed threshold to encode, this self-defined threshold value cannot ensure to be adapted to all samples, thus affects the accuracy of recognition of face.
Summary of the invention
In view of this, the invention provides a kind of face identification method and device, to solve in prior art the technical matters that the recognition of face accuracy of applying LTP operator has much room for improvement.
For solving the problem, technical scheme provided by the invention is as follows:
A kind of face identification method, described method comprises:
Determine the local three binarization mode LTP adaptive threshold eigenwert of each pixel in facial image to be identified, the corresponding LTP adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points;
According to described LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert;
Determine holotype eigenface and negative mode eigenface, described holotype eigenface is made up of each pixel holotype eigenwert, and described negative mode eigenface is made up of each pixel negative mode eigenwert;
Calculate the eigenwert histogram of described holotype eigenface and described negative mode eigenface, and calculate the information entropy weight of every layer of eigenface, utilize described information entropy weight by the eigenwert histogram weighting cascade of described holotype eigenface and described negative mode eigenface, obtain the enhancing histogram of described facial image to be identified;
Calculate the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of each known identities facial image respectively, determine recognition result according to described card side distance.
Accordingly, the described LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, comprising:
Travel through each pixel in described facial image to be identified, determine that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel in neighborhood and the gray-scale value of described current pixel point;
Calculate the standard deviation of described multiple gray scale difference value as the corresponding adaptive threshold of described current pixel point;
Using the threshold value of corresponding for described current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of described current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
Accordingly, calculate the eigenwert histogram of described holotype eigenface and described negative mode eigenface, comprising:
Calculate the quantity of holotype eigenwert described in each, obtain the eigenwert histogram of described holotype eigenface;
Calculate the quantity of negative mode eigenwert described in each, obtain the eigenwert histogram of described negative mode eigenface.
Accordingly, the information entropy weight of described calculating every layer eigenface, comprising:
Calculate holotype eigenwert described in each at the first distribution probability of described holotype eigenface and negative mode eigenwert described in each the second distribution probability in described negative mode eigenface;
Calculate the information entropy of described holotype eigenface according to described first distribution probability, calculate the information entropy of described negative mode eigenface according to described second distribution probability;
The information entropy calculating described holotype eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described holotype eigenface, the information entropy calculating described negative mode eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described negative mode eigenface.
Accordingly, describedly determine recognition result according to described card side distance, comprising:
When the enhancing histogram of described facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuse to know described facial image to be identified;
When the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as described facial image to be identified of known identities facial image corresponding to described card side distance minimum value.
Accordingly, describedly determine recognition result according to described card side distance, comprising:
When the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of the described card side known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as described facial image to be identified of one of them known identities facial image.
Accordingly, described method also comprises:
Described facial image to be identified is carried out pre-service and is divided into impartial polylith;
The described LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, comprising:
Block-by-block determines the LTP adaptive threshold eigenwert of each pixel.
A kind of face identification device, described device comprises:
First determining unit, for determining the local three binarization mode LTP adaptive threshold eigenwert of each pixel in facial image to be identified, the corresponding adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points;
Second determining unit, for according to described LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert;
3rd determining unit, for determining holotype eigenface and negative mode eigenface, described holotype eigenface is made up of each pixel holotype eigenwert, and described negative mode eigenface is made up of each pixel negative mode eigenwert;
First computing unit, for calculating the eigenwert histogram of described holotype eigenface and described negative mode eigenface;
Second computing unit, for calculating the information entropy weight of every layer of eigenface;
Obtaining unit, for utilizing described information entropy weight by the eigenwert histogram weighting cascade of described holotype eigenface and described negative mode eigenface, obtaining the enhancing histogram of described facial image to be identified;
Result recognition unit, for the enhancing histogrammic card side distance of the enhancing histogram and each known identities facial image that calculate described facial image to be identified respectively, obtains recognition result according to described card side distance.
Accordingly, described first determining unit comprises:
Gray scale difference determination subelement, for traveling through each pixel in described facial image to be identified, determines that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel in neighborhood and the gray-scale value of described current pixel point;
Threshold calculations subelement, for calculating the standard deviation of described multiple gray scale difference value as the corresponding adaptive threshold of described current pixel point;
Eigenwert determination subelement, for using the threshold value of corresponding for described current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of described current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
Accordingly, described first computing unit comprises:
First computation subunit, for calculating the quantity of holotype eigenwert described in each, obtains the eigenwert histogram of described holotype eigenface;
Second computation subunit, for calculating the quantity of negative mode eigenwert described in each, obtains the eigenwert histogram of described negative mode eigenface.
Accordingly, described second computing unit comprises:
Distribution probability computation subunit, for calculating holotype eigenwert described in each at the first distribution probability of described holotype eigenface and negative mode eigenwert described in each the second distribution probability in described negative mode eigenface;
Information entropy computation subunit, for calculating the information entropy of described holotype eigenface according to described first distribution probability, calculates the information entropy of described negative mode eigenface according to described second distribution probability;
Weight calculation subelement, information entropy for calculating described holotype eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described holotype eigenface, the information entropy calculating described negative mode eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described negative mode eigenface.
Accordingly, described result recognition unit comprises:
First recognin unit, for when the enhancing histogram of described facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuses to know described facial image to be identified;
Second recognin unit, for when the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as described facial image to be identified of known identities facial image corresponding to described card side distance minimum value.
Accordingly, described result recognition unit also comprises:
3rd recognin unit, for when the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of the described card side known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as described facial image to be identified of one of them known identities facial image.
Accordingly, described device also comprises:
Pretreatment unit, for carrying out pre-service by described facial image to be identified and being divided into impartial polylith;
Described first determining unit specifically for:
Block-by-block determines the LTP adaptive threshold eigenwert of each pixel.
As can be seen here, the present invention has following beneficial effect:
The embodiment of the present invention adopts the LTP operator with adaptive threshold to calculate the LTP adaptive threshold eigenwert of each pixel, thus obtain the eigenwert histogram of facial image to be identified, consider the light differential between image zones of different like this, adaptive threshold is calculated by neighborhood of pixel points, on the basis retaining LTP advantage, efficiently solve the Problem of Universality of LTP operator threshold value, enhance the robustness to illumination variation and noise, improve accuracy of identification.Simultaneously, information expressed by the eigenface that piece image is different is in fact different, the embodiment of the present invention is by calculating the information entropy weight of each eigenface, information entropy weight can evaluate the percentage contribution of each eigenface in feature extraction quantitatively, like this using the eigenwert histogram weighting cascade of different characteristic face as proper vector, facial image proper vector more accurately can be obtained, thus further increase the accuracy of recognition of face.
Accompanying drawing explanation
The schematic diagram of Fig. 1 (a) for utilizing LBP operator to calculate LBP eigenwert;
The schematic diagram of Fig. 1 (b) technical matters existing for LBP operator;
Fig. 2 is the schematic diagram utilizing LTP operator to calculate LTP eigenwert;
Fig. 3 is the process flow diagram of face identification method embodiment in the embodiment of the present invention;
Fig. 4 is the schematic diagram calculating LTP adaptive threshold eigenwert in the embodiment of the present invention;
Fig. 5 is the schematic diagram of holotype eigenface and negative mode eigenface in the embodiment of the present invention;
Fig. 6 is the schematic diagram of face identification device embodiment one in the embodiment of the present invention;
Fig. 7 is the schematic diagram of face identification device embodiment two in the embodiment of the present invention;
A kind of schematic diagram of terminal of Fig. 8 for providing in the embodiment of the present invention.
Embodiment
For enabling above-mentioned purpose of the present invention, feature and advantage become apparent more, are described in further detail the embodiment of the present invention below in conjunction with the drawings and specific embodiments.
For ease of the face identification method that the embodiment of the present invention provides and device are described, first the LBP operator in the face recognition algorithms related in the embodiment of the present invention and LTP operator are simply introduced.
LBP operator is the operator of Description Image textural characteristics in a kind of tonal range, is mainly used to the contrast metric of assisted extraction image local area.LBP operator is using the gray-scale value of central pixel point as threshold value, sample in center pixel neighborhood of a point, such as get the neighborhood of 3 × 3, then the gray-scale value of adjacent with central pixel point 8 pixels and threshold value are compared, if neighbor pixel gray-scale value is greater than threshold value (i.e. central pixel point gray-scale value), then this location of pixels is marked as 1, otherwise is labeled as 0.8 bits can be produced like this, 8 bits are converted to decimal number, as the LBP eigenwert of central pixel point, the decimal number span be converted to due to 8 bits is 0-255, and therefore eigenwert span is 0-255.Shown in Fig. 1 (a), give the instantiation that is asked LBP eigenwert, the grey scale pixel value of central pixel point is 9, and neighborhood territory pixel gray-scale value and center pixel gray-scale value compare, obtain 8 bits 01000111, be converted to decimal number 71 as LBP eigenwert.
But LBP operator only compares the size of gray-scale value and the contrast value that have ignored between pixel, and when the grey scale pixel value in neighborhood changes under the prerequisite keeping magnitude relationship, LBP coding result remains unchanged, shown in Fig. 1 (b).Therefore, LBP operator cannot describe the difference before and after nonlinearities change, and the important textural characteristics of part finally may be caused to be dropped.
LTP operator is the improvement to LBP operator, adopts three value codings, to improve the classification capacity of whole feature space.User Defined threshold value t, enhances the sensitivity to noise greatly, the Gao Guang that balanced to a certain extent violent illumination causes, the gray-scale value in light region.Concrete LTP operator operation process is when the difference of neighborhood territory pixel point gray-scale value and central pixel point gray-scale value is more than or equal to t, this location of pixels is marked as 1, the difference of neighborhood territory pixel point gray-scale value and central pixel point gray-scale value is less than-t, and this location of pixels is marked as-1, otherwise is labeled as 0.In order to simplify calculating, the cataloged procedure of LTP can be decomposed on the occasion of calculating section and negative value calculating section, on the occasion of with each part of negative value apply respectively LBP operator calculate method.Decomposition computation process is shown in Figure 2, and the coding result extracting "+1 " is designated as " 1 ", and all the other are designated as " 0 ", obtain holotype feature by LBP coded system; The coding result extracting "-1 " is designated as " 1 ", and all the other are designated as " 0 ", obtain negative mode feature by LBP coded system.Like this after LTP feature extraction conversion, sign and the classification performance of whole feature space sample are further enhanced and improve.But LTP adopts self-defined fixed threshold to encode, cannot ensure that self-defined threshold value is applicable to all samples, there is the Problem of Universality of threshold value.
The face identification method that the embodiment of the present invention provides and device, adopt self-defined fixed threshold for LTP operator in prior art, existence cannot be applicable to all samples, thus affects the accuracy problem of recognition of face, proposes a kind of ε-LTP adaptive thresholding algorithm.The main thought of this algorithm is the contrast value (i.e. gray scale difference value) by calculating pixel and center pixel in each local neighborhood, calculate the LTP threshold value being adaptive to this region, while not changing algorithm complex, strengthen the precision that LTP describes each local neighborhood of image.
Angle from face identification device is described by the embodiment of the present invention, this face identification device specifically can be in the client integrated, this client can be loaded in the terminal, this terminal is specifically as follows smart mobile phone, panel computer, E-book reader, MP3(Moving Picture Experts Group Audio Layer III, dynamic image expert compression standard audio frequency aspect 3) player, MP4(Moving PictureExperts Group Audio Layer IV, dynamic image expert compression standard audio frequency aspect 3) player, pocket computer on knee and desk-top computer etc.
Shown in Figure 3, be the face identification method embodiment one provided in the embodiment of the present invention, can comprise the following steps:
Step 301: the LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, the corresponding adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points.
Step 302: according to LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert.
Facial image to be identified for carrying out recognition of face adopts gray level image, first the adaptive threshold of this pixel is determined according to the gray-scale value of pixel in pixel each in facial image to be identified and this neighborhood of pixel points, recycling LTP adopts the adaptive threshold of this pixel to calculate as the threshold value of LTP operator when operator calculates the eigenwert of this pixel, namely adopts the LTP operator with adaptive threshold to determine the LTP adaptive threshold eigenwert of each pixel in facial image to be identified.
In some embodiments of the invention, determine that the specific implementation of the LTP adaptive threshold eigenwert of each pixel in facial image to be identified can comprise:
Travel through each pixel in facial image to be identified, determine that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel and the gray-scale value of current pixel point in neighborhood.
Calculate the standard deviation of multiple gray scale difference value as the corresponding adaptive threshold of current pixel point.
Using the threshold value of corresponding for current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
Also namely the embodiment of the present invention proposes a kind of adaptive threshold ε-LTP operator, extract the LTP adaptive threshold eigenwert of image, shown in Figure 4, give an instantiation so that the process calculating pixel LTP adaptive threshold eigenwert to be described in the embodiment of the present invention.General default neighborhood can get the neighborhood block of 3 × 3,8 neighbor pixels are had after then removing central pixel point, calculate the gray scale difference of these 8 neighbor pixels and central pixel point respectively, can calculate the standard deviation of this group gray scale difference as the corresponding adaptive threshold of current pixel point by this group 8 gray scale difference values of trying to achieve, recycling LTP operator calculates the LTP eigenwert of current pixel point.
In some embodiments of the invention, the face identification method that the embodiment of the present invention provides can also comprise: facial image to be identified is carried out pre-service and is divided into impartial polylith; The local three binarization mode LTP operator block-by-block with adaptive threshold can be adopted like this to calculate the LTP adaptive threshold eigenwert of each pixel, for accelerating computing velocity,
Step 303: determine holotype eigenface and negative mode eigenface, holotype eigenface is made up of each pixel holotype eigenwert, and negative mode eigenface is made up of each pixel negative mode eigenwert.
The span of pixel holotype eigenwert is 0-255, the span of negative mode eigenwert is also 0-255, like this, the gray-scale value of pixel is replaced with corresponding holotype eigenwert or negative mode eigenwert, holotype eigenface image and negative mode eigenface image can be determined respectively.Shown in Figure 5, a facial image to be identified can be converted to holotype eigenface and negative mode eigenface.
Step 304: the eigenwert histogram calculating holotype eigenface and negative mode eigenface, and calculate the information entropy weight of every layer of eigenface, utilize the information entropy weight of every layer of eigenface by the eigenwert histogram weighting cascade of holotype eigenface and negative mode eigenface, obtain the enhancing histogram of facial image to be identified.
What feature histogram calculated is the quantity of each eigenwert of piece image, in some embodiments of the invention, the histogrammic specific implementation of eigenwert calculating holotype eigenface and negative mode eigenface can comprise: the quantity calculating each holotype eigenwert, obtains the eigenwert histogram of holotype eigenface; Calculate the quantity of each negative mode eigenwert, obtain the eigenwert histogram of negative mode eigenface.
Decompose LTP feature and obtain holotype and the two-layer eigenface of negative mode normally in order to simplify calculating, but, as can be seen from Figure 5, the information that holotype eigenface and negative mode eigenface reflect is that there is a big difference, but in prior art, the gap between eigenface is not embodied by the reconstruct of proper vector, therefore, employing layered method does not have other to act on except having and simplifying calculating.For this reason, in the embodiment of the present invention, the information proposed expressed by the different characteristic layer of image is different, and information entropy therefore can be adopted to represent the quantity of information size of each layer.Carry out the percentage contribution of measures characteristic value to image expression by characteristic layer information entropy, be characteristic layer information entropy weight.Information entropy weight can evaluate the percentage contribution of each characteristic layer in feature extraction quantitatively, in multi-feature extraction and Classification and Identification, utilizes information entropy weighting to have comparatively significantly advantage.
Like this, in some embodiments of the invention, the specific implementation calculating the information entropy weight of every layer of eigenface can comprise:
Calculate each holotype eigenwert at the first distribution probability of holotype eigenface and each negative mode eigenwert the second distribution probability in negative mode eigenface;
Calculate the information entropy of holotype eigenface according to the first distribution probability, calculate the information entropy of negative mode eigenface according to the second distribution probability;
The information entropy calculating holotype eigenface accounts for the ratio of the information entropy of holotype eigenface and the information entropy sum of negative mode eigenface, as the information entropy weight of holotype eigenface, the information entropy calculating negative mode eigenface accounts for the ratio of the information entropy of holotype eigenface and the information entropy sum of negative mode eigenface, as the information entropy weight of negative mode eigenface.
By the information entropy weight of every layer of eigenface, the eigenwert histogram of holotype eigenface and negative mode eigenface is weighted cascade, the enhancing histogram obtaining facial image to be identified carries out recognition of face as the proper vector of facial image to be identified.
Step 305: calculate the enhancing histogram of facial image to be identified and the enhancing histogrammic card side distance of each known identities facial image respectively, determine recognition result according to card side's distance.
The enhancing histogram of facial image to be identified is the proper vector of facial image to be identified, compare with the enhancing histogram of the known identities facial image preserved in advance, the histogrammic acquisition pattern of enhancing of known identities facial image is consistent with the histogrammic acquisition pattern of the enhancing of facial image to be identified.
In some embodiments of the invention, determine that the specific implementation of recognition result can comprise according to card side's distance:
When the enhancing histogram of facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuse to know facial image to be identified;
When the enhancing histogram of facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as facial image to be identified of known identities facial image corresponding to card side's distance minimum value.
Namely minimum distance classifier can be utilized to classify to facial image generic to be identified.
In some embodiments of the invention, determine that the specific implementation of recognition result can also comprise according to card side's distance:
When the enhancing histogram of facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of card side's known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as facial image to be identified of one of them known identities facial image.
Namely three rank Nearest Neighbor Classifiers also can be utilized to classify to facial image generic to be identified.The thought of three rank Nearest Neighbor Classifiers is nearest three objects of selected distance, and classifications maximum belonging to choosing in three is as net result.
Like this, the embodiment of the present invention adopts the LTP operator with adaptive threshold to calculate the LTP adaptive threshold eigenwert of each pixel, thus obtain the eigenwert histogram of facial image to be identified, consider the light differential between image zones of different like this, calculate adaptive threshold by neighborhood of pixel points, on the basis retaining LTP advantage, efficiently solve the Problem of Universality of LTP operator threshold value, enhance the robustness to illumination variation and noise, improve accuracy of identification.Simultaneously, information expressed by the eigenface that piece image is different is in fact different, the embodiment of the present invention is by calculating the information entropy weight of each eigenface, information entropy weight can evaluate the percentage contribution of each eigenface in feature extraction quantitatively, like this using the eigenwert histogram weighting cascade of different characteristic face as proper vector, facial image proper vector more accurately can be obtained, thus further increase the accuracy of recognition of face.
Below by way of concrete application example and corresponding computing formula, then the face identification method that the embodiment of the present invention provides is further described.The face identification method that the embodiment of the present invention provides can be realized by following steps:
One, multilayer LTP adaptive threshold (ε-LTP) eigenface of facial image is extracted
Sample in test face database is divided into training set and test set, and being known identities facial image in training set, is facial image to be identified in test set.Such as test in face database and have 50 people, everyone each 10 width facial images, everyone gets 5 width as training set sample, another 5 width are as test set sample, from test set sample, then get arbitrarily a facial image to be identified carry out recognition of face, be same classification belonging to which the known identities facial image in this facial image to be identified of identification and training set (namely same people).
First, to all sample extraction LTP adaptive thresholds (ε-LTP) eigenface, this extracting method step is as follows:
1, to facial image piecemeal
For taking into account the local feature information of image, pretreated facial image is divided into impartial n block (value of n is chosen according to experiment effect according to the size of facial image).
2, the adaptive threshold in each neighborhood of pixel points is determined
The P neighborhood sampling of each pixel, gets 3 × 3 neighborhood blocks usually, P=8 neighbor pixel outside removing center pixel, radius of neighbourhood R=1.In neighborhood, the contrast of computing center's pixel and the interior each pixel of neighborhood (P, R), according to the characteristic of discrete values, obtains the dispersion σ of contrast value in this neighborhood:
1) central pixel point g
cin neighborhood (P, R), ask g
cwith pixel g in neighborhood
icontrast value: Δ g
i=g
i-g
c, (i=0,1 ..., P-1);
2) average of this group contrast value is calculated:
3) variance of this group contrast value is calculated:
4) dispersion (i.e. standard deviation) is calculated according to variance:
In computation process, dispersion sigma reaction is sensitive, and result can change with the change of each pixel value.Meanwhile, σ is less by the impact of sampling variation, and the dispersion of each pixel sample is more stable in the ordinary course of things, and rate of change is not high.Based on above-mentioned characteristic, divide neighborhood to ask for dispersion to pixel sample here, define the optimal threshold that this dispersion σ is this pixel LTP, namely obtain the LTP adaptive threshold ε=σ of each neighborhood of pixel self-adaptation.
3, LTP adaptive threshold (ε-LTP) eigenwert is determined
After obtaining adaptive threshold, to neighborhood point g
i(i=0,1 ..., P-1) and central pixel point g
cgray scale difference value carry out LTP coding.According to the definition of LTP operator, when the difference of neighborhood territory pixel point gray-scale value and central pixel point gray-scale value is more than or equal to t, this location of pixels is marked as 1, and the difference of neighborhood territory pixel point gray-scale value and central pixel point gray-scale value is less than-t, this location of pixels is marked as-1, otherwise is labeled as 0.Computing formula is as follows:
Wherein, g-g
cfor the gray scale difference value of neighborhood territory pixel point gray-scale value and central pixel point gray-scale value, ε is LTP adaptive threshold.
4, LTP adaptive threshold (ε-LTP) eigenwert is decomposed
Be generally to simplify and calculate, can be decomposed into positive and negative pattern when extracting LTP feature two-layer, the embodiment of the present invention adopts the method equally.After obtaining LTP adaptive threshold eigenwert, the coding result extracting "+1 " is designated as " 1 ", and all the other are designated as " 0 ", obtain holotype eigenwert by LBP coded system; The coding result extracting "-1 " is designated as " 1 ", all the other are designated as " 0 ", negative mode eigenwert is obtained by LBP coded system, after the holotype eigenwert obtaining each pixel respectively and negative mode eigenwert, the holotype eigenface be made up of the holotype eigenwert of each pixel and the negative mode eigenface be made up of the negative mode eigenwert of each pixel can be obtained further, thus obtain the eigenface of two-layer different texture.
Two, the information entropy weight of each layer LTP adaptive threshold (ε-LTP) eigenface is calculated
Information entropy is used for tolerance and determines the probabilistic quantity of information size of stochastic variable, and this uncertainty is just similar to randomness from the angle of stochastic variable, therefore calculates the stochastic distribution that information entropy is just equivalent to determine stochastic variable.In the two-layer eigenface that the self-adaptation ε-LTP shown in Fig. 5 extracts, can find out that the information expressed by the characteristic layer that image is different is different, and information entropy just can express the quantity of information size of each layer.Carry out the percentage contribution of measures characteristic value to image expression by characteristic layer information entropy, be characteristic layer information entropy weight.
When merging multilayer feature, again define information entropy according to the significance level of each layer characteristic information, namely using information entropy proportion as each layer weight, computation process is as follows:
1, the distribution probability of statistical nature layer j on eigenwert i:
wherein, the value of characteristic layer j is that 1-2, n get 256, such as, be calculate first distribution probability of each holotype eigenwert (eigenwert i span is 0-255) in holotype eigenface (j=1) during j=1
; For calculating second distribution probability of each negative mode eigenwert (eigenwert i span is 0-255) in negative mode eigenface (j=2) during j=2
2, computing information entropy E
j:
the value of same characteristic layer j is 1-2, like this can according to the first distribution probability
calculate the information entropy E of holotype eigenface
1, according to the second distribution probability
i calculates the information entropy E of negative mode eigenface
2.
3, calculate the information entropy proportion of characteristic layer j, be defined as weights W
j, characteristic layer s=2:
namely the information entropy calculating holotype eigenface accounts for the ratio of the information entropy of holotype eigenface and the information entropy sum of negative mode eigenface, as the information entropy weight of holotype eigenface, the information entropy calculating negative mode eigenface accounts for the ratio of the information entropy of holotype eigenface and the information entropy sum of negative mode eigenface, as the information entropy weight of negative mode eigenface.
Information entropy weight can evaluate the significance level of each layer feature in image characteristics extraction quantitatively, in multi-feature extraction and Classification and Identification, utilizes information entropy weighting to have comparatively significantly advantage.
Three, histogram conversion is weighted to each layer LTP adaptive threshold (ε-LTP) eigenface
Each layer LTP adaptive threshold (ε-LTP) eigenface is converted to eigenwert histogram respectively, and each layer histogram conversion regime is as follows:
wherein, m represents m value in histogram, and m gets 256,
the eigenwert quantity corresponding to each value of jth layer eigenface:
Number by adding up 1 in formula obtains
value, by all
obtain the histogram namely obtaining this layer of eigenface.If the eigenwert of pixel namely
equal with the eigenwert i value of current statistic, then I{A} is 1, otherwise is 0:
obtain the quantity that each eigenwert is corresponding.
It should be noted that, the tandem calculating eigenwert histogram and information entropy weight does not limit, the eigenwert histogram of each layer eigenface can be weighted cascade after obtaining the information entropy weight of eigenwert histogram and every layer of eigenface, form and strengthen histogram: H=(W
1h
1..., W
jh
j) (j=2), wherein, j is the number of plies of eigenface.Finally this is strengthened histogram as the proper vector describing facial image.
Four, the χ strengthened between histogram is asked
2(card side) distance, determines recognition result
1, computer card side's distance
Adopt χ
2(card side) distance function calculates the enhancing histogram of obtained facial image to be identified and the histogrammic χ of enhancing of all training sets (i.e. known identities) facial image
2distance.χ
2the computing formula of distance is as follows:
wherein, S, M represent the enhancing histogram H corresponding to two facial images that will compare respectively
i, i is sample identification.
2, with KNN Nearest Neighbor Classifier, Classification and Identification is carried out to test pattern
Three rank Nearest Neighbor Classifiers can by the χ between all histograms of obtaining
2the side's of card distance compares, and therefrom selects apart from three minimum classes, then selects optimal result wherein as facial image generic to be identified.Utilize three rank nearest neighbour methods to calculate with test pattern apart from three minimum width images, classifications maximum belonging to choosing in three, as net result, can obtain the recognition result of facial image to be identified thus.
Like this, the embodiment of the present invention adopts the LTP operator with adaptive threshold to calculate the LTP adaptive threshold eigenwert of each pixel, thus obtain the eigenwert histogram of facial image to be identified, consider the light differential between image zones of different like this, calculate adaptive threshold by neighborhood of pixel points, on the basis retaining LTP advantage, efficiently solve the Problem of Universality of LTP operator threshold value, enhance the robustness to illumination variation and noise, improve accuracy of identification.Simultaneously, information expressed by the eigenface that piece image is different is in fact different, the embodiment of the present invention is by calculating the information entropy weight of each eigenface, information entropy weight can evaluate the percentage contribution of each eigenface in feature extraction quantitatively, like this using the eigenwert histogram weighting cascade of different characteristic face as proper vector, facial image proper vector more accurately can be obtained, thus further increase the accuracy of recognition of face.
Shown in Figure 6, be face identification device embodiment one a kind of in the embodiment of the present invention, this device can comprise:
First determining unit 601, for determining the LTP adaptive threshold eigenwert of each pixel in facial image to be identified, the corresponding adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points;
Second determining unit 602, for according to LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert;
3rd determining unit 603, for determining holotype eigenface and negative mode eigenface, holotype eigenface is made up of each pixel holotype eigenwert, and negative mode eigenface is made up of each pixel negative mode eigenwert;
First computing unit 604, for calculating the eigenwert histogram of holotype eigenface and negative mode eigenface;
Second computing unit 605, for calculating the information entropy weight of every layer of eigenface;
Obtaining unit 606, for utilizing information entropy weight by the eigenwert histogram weighting cascade of holotype eigenface and negative mode eigenface, obtaining the enhancing histogram of facial image to be identified;
Result recognition unit 607, for the enhancing histogrammic card side distance of the enhancing histogram and each known identities facial image that calculate facial image to be identified respectively, obtains recognition result according to card side's distance.
Shown in Figure 7, be face identification device embodiment two a kind of in the embodiment of the present invention, in some embodiments of the invention, the first determining unit 601 can comprise:
Gray scale difference determination subelement 6011, for traveling through each pixel in facial image to be identified, determines that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel and the gray-scale value of current pixel point in neighborhood;
Threshold calculations subelement 6012, for calculating the standard deviation of multiple gray scale difference value as the corresponding adaptive threshold of current pixel point;
Eigenwert determination subelement 6013, for using the threshold value of corresponding for current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
In some embodiments of the invention, the first computing unit 604 can comprise:
First computation subunit 6041, for calculating the quantity of each holotype eigenwert, obtains the eigenwert histogram of holotype eigenface;
Second computation subunit 6042, for calculating the quantity of each negative mode eigenwert, obtains the eigenwert histogram of negative mode eigenface.
In some embodiments of the invention, the second computing unit 605 can comprise:
Distribution probability computation subunit 6051, for calculating each holotype eigenwert at the first distribution probability of holotype eigenface and each negative mode eigenwert the second distribution probability in negative mode eigenface;
Information entropy computation subunit 6052, for calculating the information entropy of holotype eigenface according to the first distribution probability, calculates the information entropy of negative mode eigenface according to the second distribution probability;
Weight calculation subelement 6053, information entropy for calculating holotype eigenface accounts for the ratio of the information entropy of holotype eigenface and the information entropy sum of negative mode eigenface, as the information entropy weight of holotype eigenface, the information entropy calculating negative mode eigenface accounts for the ratio of the information entropy of holotype eigenface and the information entropy sum of negative mode eigenface, as the information entropy weight of negative mode eigenface.
In some embodiments of the invention, result recognition unit can comprise:
First recognin unit, for when the enhancing histogram of facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuses to know facial image to be identified;
Second recognin unit, for when the enhancing histogram of facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as facial image to be identified of known identities facial image corresponding to card side's distance minimum value.
In some embodiments of the invention, result recognition unit can also comprise:
3rd recognin unit, for when the enhancing histogram of facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of card side's known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as facial image to be identified of one of them known identities facial image.
In some embodiments of the invention, the face identification device that the embodiment of the present invention provides can also comprise:
Pretreatment unit, for carrying out pre-service by facial image to be identified and being divided into impartial polylith;
Then the first determining unit specifically for:
Block-by-block determines the LTP adaptive threshold eigenwert of each pixel.
Like this, the embodiment of the present invention adopts the LTP operator with adaptive threshold to calculate the LTP adaptive threshold eigenwert of each pixel, thus obtain the eigenwert histogram of facial image to be identified, consider the light differential between image zones of different like this, calculate adaptive threshold by neighborhood of pixel points, on the basis retaining LTP advantage, efficiently solve the Problem of Universality of LTP operator threshold value, enhance the robustness to illumination variation and noise, improve accuracy of identification.Simultaneously, information expressed by the eigenface that piece image is different is in fact different, the embodiment of the present invention is by calculating the information entropy weight of each eigenface, information entropy weight can evaluate the percentage contribution of each eigenface in feature extraction quantitatively, like this using the eigenwert histogram weighting cascade of different characteristic face as proper vector, facial image proper vector more accurately can be obtained, thus further increase the accuracy of recognition of face.
Accordingly, the embodiment of the present invention also provides a kind of terminal, shown in Figure 8, can comprise:
Processor 801, storer 802, input media 803 and output unit 804.The quantity of the processor 801 in browser server can be one or more, for a processor in Fig. 8.In some embodiments of the invention, processor 801, storer 802, input media 803 are connected by bus or alternate manner with output unit 804, wherein, to be connected by bus in Fig. 8.
Storer 802 can be used for storing software program and module, and processor 801 is stored in software program and the module of storer 802 by running, thus performs various function application and the data processing of browser server.Storer 802 mainly can comprise storage program district and store data field, and wherein, storage program district can store operating system, application program etc. needed at least one function.In addition, storer 802 can comprise high-speed random access memory, can also comprise nonvolatile memory, such as at least one disk memory, flush memory device or other volatile solid-state parts.Input media 803 can be used for the numeral or the character information that receive input, and generation arranges with the user of browser server and function controls the input of relevant key signals.
Specifically in the present embodiment, processor 801 can according to following instruction, executable file corresponding for the process of one or more application program is loaded in storer 802, and is run the application program be stored in storer 802 by processor 801, thus realize various function:
Determine the local three binarization mode LTP adaptive threshold eigenwert of each pixel in facial image to be identified, the corresponding LTP adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points;
According to described LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert;
Determine holotype eigenface and negative mode eigenface, described holotype eigenface is made up of each pixel holotype eigenwert, and described negative mode eigenface is made up of each pixel negative mode eigenwert;
Calculate the eigenwert histogram of described holotype eigenface and described negative mode eigenface, and calculate the information entropy weight of every layer of eigenface, utilize described information entropy weight by the eigenwert histogram weighting cascade of described holotype eigenface and described negative mode eigenface, obtain the enhancing histogram of described facial image to be identified;
Calculate the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of each known identities facial image respectively, determine recognition result according to described card side distance.
Accordingly, the described LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, comprising:
Travel through each pixel in described facial image to be identified, determine that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel in neighborhood and the gray-scale value of described current pixel point;
Calculate the standard deviation of described multiple gray scale difference value as the corresponding adaptive threshold of described current pixel point;
Using the threshold value of corresponding for described current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of described current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
Accordingly, calculate the eigenwert histogram of described holotype eigenface and described negative mode eigenface, comprising:
Calculate the quantity of holotype eigenwert described in each, obtain the eigenwert histogram of described holotype eigenface;
Calculate the quantity of negative mode eigenwert described in each, obtain the eigenwert histogram of described negative mode eigenface.
Accordingly, the information entropy weight of described calculating every layer eigenface, comprising:
Calculate holotype eigenwert described in each at the first distribution probability of described holotype eigenface and negative mode eigenwert described in each the second distribution probability in described negative mode eigenface;
Calculate the information entropy of described holotype eigenface according to described first distribution probability, calculate the information entropy of described negative mode eigenface according to described second distribution probability;
The information entropy calculating described holotype eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described holotype eigenface, the information entropy calculating described negative mode eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described negative mode eigenface.
Accordingly, describedly determine recognition result according to described card side distance, comprising:
When the enhancing histogram of described facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuse to know described facial image to be identified;
When the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as described facial image to be identified of known identities facial image corresponding to described card side distance minimum value.
Accordingly, describedly determine recognition result according to described card side distance, comprising:
When the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of the described card side known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as described facial image to be identified of one of them known identities facial image.
Accordingly, also comprise:
Described facial image to be identified is carried out pre-service and is divided into impartial polylith;
The described LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, comprising:
Block-by-block determines the LTP adaptive threshold eigenwert of each pixel.
Like this, the embodiment of the present invention adopts the LTP operator with adaptive threshold to calculate the LTP adaptive threshold eigenwert of each pixel, thus obtain the eigenwert histogram of facial image to be identified, consider the light differential between image zones of different like this, calculate adaptive threshold by neighborhood of pixel points, on the basis retaining LTP advantage, efficiently solve the Problem of Universality of LTP operator threshold value, enhance the robustness to illumination variation and noise, improve accuracy of identification.Simultaneously, information expressed by the eigenface that piece image is different is in fact different, the embodiment of the present invention is by calculating the information entropy weight of each eigenface, information entropy weight can evaluate the percentage contribution of each eigenface in feature extraction quantitatively, like this using the eigenwert histogram weighting cascade of different characteristic face as proper vector, facial image proper vector more accurately can be obtained, thus further increase the accuracy of recognition of face.
It should be noted that, in this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is the difference with other embodiments, between each embodiment identical similar portion mutually see.For system disclosed in embodiment or device, because it corresponds to the method disclosed in Example, so description is fairly simple, relevant part illustrates see method part.
Also it should be noted that, in this article, the such as relational terms of first and second grades and so on is only used for an entity or operation to separate with another entity or operational zone, and not necessarily requires or imply the relation that there is any this reality between these entities or operation or sequentially.And, term " comprises ", " comprising " or its any other variant are intended to contain comprising of nonexcludability, thus make to comprise the process of a series of key element, method, article or equipment and not only comprise those key elements, but also comprise other key elements clearly do not listed, or also comprise by the intrinsic key element of this process, method, article or equipment.When not more restrictions, the key element limited by statement " comprising ... ", and be not precluded within process, method, article or the equipment comprising described key element and also there is other identical element.
The software module that the method described in conjunction with embodiment disclosed herein or the step of algorithm can directly use hardware, processor to perform, or the combination of the two is implemented.Software module can be placed in the storage medium of other form any known in random access memory (RAM), internal memory, ROM (read-only memory) (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field.
To the above-mentioned explanation of the disclosed embodiments, professional and technical personnel in the field are realized or uses the present invention.To be apparent for those skilled in the art to the multiple amendment of these embodiments, General Principle as defined herein can without departing from the spirit or scope of the present invention, realize in other embodiments.Therefore, the present invention can not be restricted to these embodiments shown in this article, but will meet the widest scope consistent with principle disclosed herein and features of novelty.
Claims (14)
1. a face identification method, is characterized in that, described method comprises:
Determine the local three binarization mode LTP adaptive threshold eigenwert of each pixel in facial image to be identified, the corresponding LTP adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points;
According to described LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert;
Determine holotype eigenface and negative mode eigenface, described holotype eigenface is made up of each pixel holotype eigenwert, and described negative mode eigenface is made up of each pixel negative mode eigenwert;
Calculate the eigenwert histogram of described holotype eigenface and described negative mode eigenface, and calculate the information entropy weight of every layer of eigenface, utilize described information entropy weight by the eigenwert histogram weighting cascade of described holotype eigenface and described negative mode eigenface, obtain the enhancing histogram of described facial image to be identified;
Calculate the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of each known identities facial image respectively, determine recognition result according to described card side distance.
2. method according to claim 1, is characterized in that, the described LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, comprising:
Travel through each pixel in described facial image to be identified, determine that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel in neighborhood and the gray-scale value of described current pixel point;
Calculate the standard deviation of described multiple gray scale difference value as the corresponding adaptive threshold of described current pixel point;
Using the threshold value of corresponding for described current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of described current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
3. method according to claim 1, is characterized in that, calculates the eigenwert histogram of described holotype eigenface and described negative mode eigenface, comprising:
Calculate the quantity of holotype eigenwert described in each, obtain the eigenwert histogram of described holotype eigenface;
Calculate the quantity of negative mode eigenwert described in each, obtain the eigenwert histogram of described negative mode eigenface.
4. method according to claim 1, is characterized in that, the information entropy weight of described calculating every layer eigenface, comprising:
Calculate holotype eigenwert described in each at the first distribution probability of described holotype eigenface and negative mode eigenwert described in each the second distribution probability in described negative mode eigenface;
Calculate the information entropy of described holotype eigenface according to described first distribution probability, calculate the information entropy of described negative mode eigenface according to described second distribution probability;
The information entropy calculating described holotype eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described holotype eigenface, the information entropy calculating described negative mode eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described negative mode eigenface.
5. method according to claim 1, is characterized in that, describedly determines recognition result according to described card side distance, comprising:
When the enhancing histogram of described facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuse to know described facial image to be identified;
When the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as described facial image to be identified of known identities facial image corresponding to described card side distance minimum value.
6. method according to claim 5, is characterized in that, describedly determines recognition result according to described card side distance, comprising:
When the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of the described card side known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as described facial image to be identified of one of them known identities facial image.
7. method according to claim 1, is characterized in that, described method also comprises:
Described facial image to be identified is carried out pre-service and is divided into impartial polylith;
The described LTP adaptive threshold eigenwert determining each pixel in facial image to be identified, comprising:
Block-by-block determines the LTP adaptive threshold eigenwert of each pixel.
8. a face identification device, is characterized in that, described device comprises:
First determining unit, for determining the local three binarization mode LTP adaptive threshold eigenwert of each pixel in facial image to be identified, the corresponding adaptive threshold of each pixel is determined by the gray scale difference value of each pixel in this pixel and this neighborhood of pixel points;
Second determining unit, for according to described LTP adaptive threshold eigenwert determination holotype eigenwert and negative mode eigenwert;
3rd determining unit, for determining holotype eigenface and negative mode eigenface, described holotype eigenface is made up of each pixel holotype eigenwert, and described negative mode eigenface is made up of each pixel negative mode eigenwert;
First computing unit, for calculating the eigenwert histogram of described holotype eigenface and described negative mode eigenface;
Second computing unit, for calculating the information entropy weight of every layer of eigenface;
Obtaining unit, for utilizing described information entropy weight by the eigenwert histogram weighting cascade of described holotype eigenface and described negative mode eigenface, obtaining the enhancing histogram of described facial image to be identified;
Result recognition unit, for the enhancing histogrammic card side distance of the enhancing histogram and each known identities facial image that calculate described facial image to be identified respectively, obtains recognition result according to described card side distance.
9. device according to claim 8, is characterized in that, described first determining unit comprises:
Gray scale difference determination subelement, for traveling through each pixel in described facial image to be identified, determines that the current pixel point traversed presets the gray scale difference of the gray-scale value of each pixel in neighborhood and the gray-scale value of described current pixel point;
Threshold calculations subelement, for calculating the standard deviation of described multiple gray scale difference value as the corresponding adaptive threshold of described current pixel point;
Eigenwert determination subelement, for using the threshold value of corresponding for described current pixel point adaptive threshold as LTP operator, adopt the LTP eigenwert with the LTP operator determination current pixel point of adaptive threshold, the LTP eigenwert of described current pixel point is the LTP adaptive threshold eigenwert of current pixel point.
10. device according to claim 8, is characterized in that, described first computing unit comprises:
First computation subunit, for calculating the quantity of holotype eigenwert described in each, obtains the eigenwert histogram of described holotype eigenface;
Second computation subunit, for calculating the quantity of negative mode eigenwert described in each, obtains the eigenwert histogram of described negative mode eigenface.
11. devices according to claim 8, is characterized in that, described second computing unit comprises:
Distribution probability computation subunit, for calculating holotype eigenwert described in each at the first distribution probability of described holotype eigenface and negative mode eigenwert described in each the second distribution probability in described negative mode eigenface;
Information entropy computation subunit, for calculating the information entropy of described holotype eigenface according to described first distribution probability, calculates the information entropy of described negative mode eigenface according to described second distribution probability;
Weight calculation subelement, information entropy for calculating described holotype eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described holotype eigenface, the information entropy calculating described negative mode eigenface accounts for the ratio of the information entropy of described holotype eigenface and the information entropy sum of described negative mode eigenface, as the information entropy weight of described negative mode eigenface.
12. devices according to claim 8, is characterized in that, described result recognition unit comprises:
First recognin unit, for when the enhancing histogram of described facial image to be identified and the enhancing of each known identities facial image histogrammic card side distance are all greater than predetermined threshold value, then refuses to know described facial image to be identified;
Second recognin unit, for when the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least one known identities facial image are less than or equal to predetermined threshold value, choose the classification of classification as described facial image to be identified of known identities facial image corresponding to described card side distance minimum value.
13. devices according to claim 12, is characterized in that, described result recognition unit also comprises:
3rd recognin unit, for when the enhancing histogram of described facial image to be identified and the enhancing histogrammic card side distance of at least three known identities facial images are less than or equal to predetermined threshold value, utilize three rank Nearest Neighbor Classifiers from the classification of the described card side known identities facial image corresponding apart from minimum three numerical value, choose the classification of classification as described facial image to be identified of one of them known identities facial image.
14. devices according to claim 8, is characterized in that, described device also comprises:
Pretreatment unit, for carrying out pre-service by described facial image to be identified and being divided into impartial polylith;
Described first determining unit specifically for:
Block-by-block determines the LTP adaptive threshold eigenwert of each pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410088084.9A CN104915626A (en) | 2014-03-11 | 2014-03-11 | Face identification method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410088084.9A CN104915626A (en) | 2014-03-11 | 2014-03-11 | Face identification method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104915626A true CN104915626A (en) | 2015-09-16 |
Family
ID=54084684
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410088084.9A Pending CN104915626A (en) | 2014-03-11 | 2014-03-11 | Face identification method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104915626A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106649611A (en) * | 2016-11-29 | 2017-05-10 | 哈尔滨商业大学 | Image retrieval method based on neighborhood rotation right angle mode |
| CN106778529A (en) * | 2016-11-25 | 2017-05-31 | 南京理工大学 | A kind of face identification method based on improvement LDP |
| CN107038699A (en) * | 2016-11-09 | 2017-08-11 | 重庆医科大学 | Strengthen image fault rate detection method |
| CN107346410A (en) * | 2016-05-05 | 2017-11-14 | 杭州海康威视数字技术股份有限公司 | image processing method and device |
| CN109978846A (en) * | 2019-03-18 | 2019-07-05 | 哈尔滨商业大学 | A kind of Lung neoplasm texture feature extraction system and method based on three value mode of body local direction |
| CN111461147A (en) * | 2020-04-30 | 2020-07-28 | 柳州智视科技有限公司 | Binary coding organization algorithm based on image characteristics |
| CN112183394A (en) * | 2020-09-30 | 2021-01-05 | 江苏智库智能科技有限公司 | Face recognition method and device and intelligent security management system |
| CN112488121A (en) * | 2020-11-16 | 2021-03-12 | 广州杰赛科技股份有限公司 | Vehicle identification method, device, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102163283A (en) * | 2011-05-25 | 2011-08-24 | 电子科技大学 | Method for extracting face characteristic based on local three-value mode |
| CN103778412A (en) * | 2014-01-16 | 2014-05-07 | 重庆邮电大学 | Face recognition method based on local ternary pattern adaptive threshold |
-
2014
- 2014-03-11 CN CN201410088084.9A patent/CN104915626A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102163283A (en) * | 2011-05-25 | 2011-08-24 | 电子科技大学 | Method for extracting face characteristic based on local three-value mode |
| CN103778412A (en) * | 2014-01-16 | 2014-05-07 | 重庆邮电大学 | Face recognition method based on local ternary pattern adaptive threshold |
Non-Patent Citations (2)
| Title |
|---|
| M. AKHLOUFI ET AL.: ""A new fusion framework for multispectral face recognition in the texture space"", 《10TH INTERNATIONAL CONFERENCE ON QUANTITATIVE INFRARED THERMOGRAPHY》 * |
| 周丽芳 等: ""一种自适应的EDTLBP人脸识别方法"", 《重庆邮电大学学报（自然科学版）》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107346410A (en) * | 2016-05-05 | 2017-11-14 | 杭州海康威视数字技术股份有限公司 | image processing method and device |
| CN107346410B (en) * | 2016-05-05 | 2020-03-06 | 杭州海康威视数字技术股份有限公司 | Picture processing method and device |
| CN107038699A (en) * | 2016-11-09 | 2017-08-11 | 重庆医科大学 | Strengthen image fault rate detection method |
| CN107038699B (en) * | 2016-11-09 | 2019-07-23 | 重庆医科大学 | Enhance image fault rate detection method |
| CN106778529A (en) * | 2016-11-25 | 2017-05-31 | 南京理工大学 | A kind of face identification method based on improvement LDP |
| CN106649611A (en) * | 2016-11-29 | 2017-05-10 | 哈尔滨商业大学 | Image retrieval method based on neighborhood rotation right angle mode |
| CN106649611B (en) * | 2016-11-29 | 2020-04-24 | 哈尔滨商业大学 | Image retrieval method based on neighborhood rotation right-angle mode |
| CN109978846A (en) * | 2019-03-18 | 2019-07-05 | 哈尔滨商业大学 | A kind of Lung neoplasm texture feature extraction system and method based on three value mode of body local direction |
| CN111461147A (en) * | 2020-04-30 | 2020-07-28 | 柳州智视科技有限公司 | Binary coding organization algorithm based on image characteristics |
| CN112183394A (en) * | 2020-09-30 | 2021-01-05 | 江苏智库智能科技有限公司 | Face recognition method and device and intelligent security management system |
| CN112488121A (en) * | 2020-11-16 | 2021-03-12 | 广州杰赛科技股份有限公司 | Vehicle identification method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104915626A (en) | Face identification method and apparatus | |
| Cheng et al. | RetinaNet with difference channel attention and adaptively spatial feature fusion for steel surface defect detection | |
| Li et al. | Efficient densely connected convolutional neural networks | |
| Wu et al. | Improved k-nearest neighbor classification | |
| Liao | Region description using extended local ternary patterns | |
| CN104778457B (en) | Video face identification method based on multi-instance learning | |
| Basheer et al. | FESD: An approach for biometric human footprint matching using fuzzy ensemble learning | |
| CN103778412A (en) | Face recognition method based on local ternary pattern adaptive threshold | |
| CN104881871A (en) | Traffic image segmentation method based on improved multi-object harmony search algorithm | |
| CN102955855A (en) | Palm print database search method based on quantum algorithms | |
| CN104881449A (en) | Image retrieval method based on manifold learning data compression hash | |
| CN101964061B (en) | Binary kernel function support vector machine-based vehicle type recognition method | |
| Zhou et al. | Fast minutiae extractor using neural network | |
| Jampour et al. | Chaos game theory and its application for offline signature identification | |
| Singh et al. | Feature selection based classifier combination approach for handwritten Devanagari numeral recognition | |
| Luo et al. | Texture classification combining improved local binary pattern and threshold segmentation | |
| CN111428701A (en) | Small-area fingerprint image feature extraction method, system, terminal and storage medium | |
| Liang et al. | Adaptive multiple kernel fusion model using spatial-statistical information for high resolution SAR image classification | |
| Loka et al. | Long range correlation of preceded pixels relations and application to off‐line signature verification | |
| Lv et al. | An algorithm of Iris feature-extracting based on 2D Log-Gabor | |
| CN102129561A (en) | Fingerprint image segmentation method based on artificial immunogenetics and morphology | |
| CN109359607A (en) | A kind of palm print and palm vein fusion identification method based on texture | |
| Parasnis et al. | RoadScan: A Novel and Robust Transfer Learning Framework for Autonomous Pothole Detection in Roads | |
| Qin et al. | Multi-scaling detection of singular points based on fully convolutional networks in fingerprint images | |
| Le et al. | Ent-Boost: Boosting using entropy measures for robust object detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150916 |
|
| RJ01 | Rejection of invention patent application after publication |