CN105095836B - A kind of skin-texture detection method and device based on Gabor characteristic - Google Patents
A kind of skin-texture detection method and device based on Gabor characteristic Download PDFInfo
- Publication number
- CN105095836B CN105095836B CN201410203716.1A CN201410203716A CN105095836B CN 105095836 B CN105095836 B CN 105095836B CN 201410203716 A CN201410203716 A CN 201410203716A CN 105095836 B CN105095836 B CN 105095836B
- Authority
- CN
- China
- Prior art keywords
- classifier
- image block
- training sample
- image
- skin
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention is suitable for technical field of image processing, provides a kind of skin-texture detection method and device based on Gabor characteristic, which comprises image to be detected is divided into multiple images block；Each color channel of each image block is successively done into convolution in airspace with the Gabor filter group pre-established, obtains k*s*r matrixes identical with the image block size size；The mean value and variance of each matrix are calculated, and using the mean value and variance as characteristic value, obtains k*s*r*2 characteristic value；Using the k*s*r*2 characteristic value as the feature vector of the image block, and described eigenvector is passed through into the classifier pre-established and is detected, to determine whether the image block has dermatoglyph feature.The image with dermatoglyph feature can be fast and accurately detected through the invention.
Description
Technical field
The invention belongs to technical field of image processing more particularly to a kind of skin-texture detection sides based on Gabor characteristic
Method and device.
Background technique
In recent years, image detection received significant attention, due to including a large amount of texture information, and different parts in image
Textural characteristics differ greatly.Therefore, texture has received widespread attention as important low-level visual feature, and skin texture detection is
Through becoming important field of research in image detection.
With the fast development of internet, network also produces many bad letters while bringing convenient
Breath, especially porny produces adverse effect to teenager.In the porny detection based on the colour of skin, sandy beach, wood
Line etc. is easy to misjudged with object similar in skin color, but these can be easy to be filtered by detections of texture.
Existing texture detection mainly has statistical method, and statistical method includes gray level co-occurrence matrixes, part ash
The methods of degree statistics.But existing gray level co-occurrence matrixes, local gray level statistics the methods of when carrying out skin texture detection, in occupancy
Deposit that more, the speed of service is slow, detection effect is poor.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of skin-texture detection method and device based on Gabor characteristic,
To solve the prior art when carrying out skin texture detection, existing committed memory is more, the speed of service is slow and detection effect is poor
Problem.
The embodiments of the present invention are implemented as follows, a kind of skin-texture detection method based on Gabor characteristic, the method
Include:
Image to be detected is divided into multiple images block；
Each color channel of each image block is successively rolled up in airspace with the Gabor filter group pre-established
Product obtains k*s*r matrixes identical with the image block size size, and wherein k is in the direction of the Gabor filter group
Number, s are the number of the scale of the Gabor filter group, and r is the number of the color channel, and k, s, r are whole greater than zero
Number；
The mean value and variance of each matrix are calculated, and using the mean value and variance as characteristic value, obtains k*s*
R*2 characteristic value；
Using the k*s*r*2 characteristic value as the feature vector of the image block, and described eigenvector is passed through in advance
The classifier of foundation is detected, to determine whether the image block has dermatoglyph feature.
The another object of the embodiment of the present invention is to provide a kind of skin-texture detection device based on Gabor characteristic, institute
Stating device includes:
First image block division unit, for image to be detected to be divided into multiple images block；
First computing unit, for filtering each color channel of each image block with the Gabor pre-established
Device group successively does convolution in airspace, obtains k*s*r matrixes identical with the image block size size, and wherein k is the Gabor
The number in the direction of filter group, s are the number of the scale of the Gabor filter group, and r is the number of the color channel,
K, s, r are the integer greater than zero；
Second computing unit is made for calculating the mean value and variance of each matrix, and by the mean value and variance
It is characterized value, obtains k*s*r*2 characteristic value；
Detection unit, for using the k*s*r*2 characteristic value as the feature vector of the image block, and by the feature
Vector passes through the classifier pre-established and is detected, to determine whether the image block has dermatoglyph feature.
Existing beneficial effect is the embodiment of the present invention compared with prior art: the embodiment of the present invention is by by mapping to be checked
As being divided into multiple images block, each color channel of each image block and the Gabor filter group of foundation are carried out small
Wave conversion, and transformation results are extracted into mean value and variance as characteristic value, so that improving the same of dermatoglyph judging nicety rate
When reduce the Spatial Dimension of feature vector, save memory headroom.Since the transformation results are identical as tile size
Matrix, if the Spatial Dimension of feature vector can be very big, therefore the present invention is real by the transformation results directly as characteristic value
Example is applied by the mean value of matrix and variance as characteristic value, greatly reduces Spatial Dimension while retaining main feature.Moreover,
The embodiment of the present invention is additionally contemplates that the color of skin is relatively simple, and the component on RGB different color channel extracts respectively,
So that the extraction accuracy rate than gray level image is higher by very much.In addition, the embodiment of the present invention can choose multiple directions and multiple rulers
The Gabor filter group of degree is allowed to be easier to coincide with the direction of image block and scale, so that energy is larger after wavelet transformation,
Better effect can be obtained in skin texture analysis.Skin-texture detection of embodiment of the present invention accuracy rate is high, speed is fast, is easy to
Realize, and do not need to increase additional hardware during entire skin-texture detection, so as to effectively reduce cost, have compared with
Strong usability and practicality.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description is only of the invention some
Embodiment for those of ordinary skill in the art without any creative labor, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1 is the implementation flow chart for the skin-texture detection method based on Gabor characteristic that the embodiment of the present invention one provides；
2a, 2b in Fig. 2 are showing for the skin-texture detection effect based on Gabor characteristic that the embodiment of the present invention one provides
Example diagram；
Fig. 3 is the composite structural diagram of the skin-texture detection device provided by Embodiment 2 of the present invention based on Gabor characteristic.
Specific embodiment
In being described below, for illustration and not for limitation, the tool of such as particular system structure, technology etc is proposed
Body details understands the embodiment of the present invention to cut thoroughly.However, it will be clear to one skilled in the art that there is no these specific
The present invention also may be implemented in the other embodiments of details.In other situations, it omits to well-known system, device, electricity
The detailed description of road and method, in case unnecessary details interferes description of the invention.
In order to illustrate technical solutions according to the invention, the following is a description of specific embodiments.
Embodiment one:
Fig. 1 shows the realization stream of the skin-texture detection method based on Gabor characteristic of the offer of the embodiment of the present invention one
Journey, details are as follows for this method process:
In step s101, image to be detected is divided into multiple images block.
Illustratively, described image to be detected can be divided into the image block of 16*16 size by the present embodiment.
It should be noted that since textural characteristics are the features of a kind of dependence and image peripheral pixel relationship, so, if
Image block is too small, can not embody the feature that this circulation occurs, and calculation amount can greatly increase；If image block is too big, one
A image block may both comprising dermatoglyph feature or include non-skin textural characteristics, can reduce the standard of dermatoglyph feature detection
True property.For the present embodiment after by specifically testing, the image block of preferably 16*16 pixel divides image.
In step s 102, by each color channel of each image block and the Gabor filter group pre-established
Convolution successively is done in airspace, obtains k*s*r matrixes identical with the image block size size, wherein k is Gabor filtering
The number in the direction of device group, s are the number of the scale of the Gabor filter group, and r is the number of the color channel, k, s, r
For the integer greater than zero.
In view of the color of skin is relatively simple, in order to improve the accuracy rate of extraction, the present embodiment is logical in RGB different color
Component on road extracts respectively, i.e., by the Gabor filter group of each color channel of each image block and foundation
Carry out wavelet transformation.
Wherein, the process for establishing Gabor filter group is as follows:
Gabor function is the Gaussian function of multiple sine curve modulation, its two-dimentional expression formula is:
Gabor filter group function can pass through flexible and translation g (x, y) Lai Shengcheng: gmn(x, y)=a-mg(x',y'),a>
1, m, n is integer；
Wherein: x'=a-m(xcos θ+ysin θ), y'=a-m(- xsin θ+ycos θ), θ=n π/k
Parameter a, σuAnd σvCalculation formula it is as follows:
K is the number in Gabor filter group direction, and s is the number of Gabor filter group scale, high ulAnd uhIt is low respectively
The centre frequency of frequency and high frequency, scale factor a-mIt ensure that energy independently of m.Illustratively, uh=0.4, ul=0.05, k=
6, s=4.
The embodiment of the present invention carries out small echo to image block by selection multiple directions and the Gabor filter group of multiple scales
Transformation is allowed to be easier to coincide with the direction of image block and scale, so that energy is larger after wavelet transformation, in skin texture analysis
On can obtain better effect.
It should be noted that the present embodiment has used the Gabor in signal processing method in dermatoglyph feature extraction
Wavelet transformation extracts.Wavelet analysis is a kind of Time-Frequency Analysis Method, can verify the localized variation of signal or image, and Gabor
Transformation has the smallest time frequency window, and Gabor function and the visual sense feeling of mammal also quite coincide.Gabor wavelet collection
Nonorthogonality cause filtering after image have redundancy.During designing filter, by setting low frequency ulWith
High frequency uhCenter, to ensure that the response of filter group can contact with each other in frequency upper half peak amplitude, and do not overlap each other.It is inciting somebody to action
Image is transformed into after frequency domain, and different texture has different centre frequency and bandwidth, goes out one group of filter by frequency and bandwidth Design
Wave device, each filter only allow corresponding texture to pass through, and the energy of other textures will receive inhibition.When grain direction with
When filter direction is more identical, output energy is bigger.The present embodiment is after by specifically testing, preferably 4 scales and 6
The filter group in direction.
In step s 103, the mean value and variance of each matrix are calculated, and using the mean value and variance as spy
Value indicative obtains k*s*r*2 characteristic value.
In the present embodiment, since the Gabor filter group transformation results are identical with the size of image block
Matrix, if the Spatial Dimension of feature vector can be very big, therefore the present invention is implemented directly using the transformation results as characteristic value
The mean value of matrix and variance as characteristic value, are greatly reduced Spatial Dimension while retaining main feature by example.
With above-mentioned k=6, for s=4, it is assumed that certain image block areas is I (x, y), then can calculate:
Wks(x, y)=∫ I (x1,y1)gks(x-x1,y-y1)dx1dy1
Wherein W is function I's and g's convolution, and g is filter group function, and I is that image-region function is (really discrete
Value).
The mean μ and variances sigma of convolution coefficient2It is as follows by function representation for final characteristic value:
μks=∫ ∫ | Wks(x,y)|dxdy
What is be achieved in that is that this matrix is extracted mean μ and variances sigma with the identical matrix of image block size size2,
Then form the characteristic value of the direction k s scale.Each the color dimension for successively calculating each image block can obtain 4*6=24 group spy
Value indicative.Ultimately form the feature vector of 24*3*2 dimension.
In step S104, using the m*n*s*2 characteristic value as the feature vector of the image block, and by the feature
Vector passes through the classifier pre-established and is detected, to determine whether the image block has dermatoglyph feature.
In the present embodiment, it establishes classifier detailed process is as follows:
A, multiple training images are acquired, the training image includes dermatoglyph feature.
In the present embodiment, the image for multiple (such as 100) passing through colour of skin processing is acquired, due to dermatoglyph to be trained,
So including the baring skin of large area in treated image.
B, the training image is divided into multiple images block.
Illustratively, the image for 512*512 is uniformly processed in the size of the training image, then successively carries out piecemeal,
The size of image block is 16*16.
C, the feature vector of each image block is obtained, described eigenvector includes multiple characteristic values.
The spy comprising multiple characteristic values is obtained using above-mentioned (i.e. step S101~step S104) mode to each image block
Vector is levied, as detailed above, details are not described herein.
D, according to the characteristic value, described image block is clustered using gauss hybrid models, obtains training sample, and
The training sample is labeled as two class of skin and non-skin.
Since gauss hybrid models (Gaussian Mixture Model, GMM) being capable of smoothly approximate arbitrary shape
Density Distribution can obtain good effect in image recognition.Therefore the present embodiment gathers described image block using GMM
Class, and initial clustering point is selected using k-means algorithms selection, GMM fast convergence can be made.
Details are as follows for detailed process:
One group of characteristic value regarded to the point of hyperspace as, dimension is d, number of training n, distributions of these points can be with
It is indicated with the weighted average of multiple Gaussian functions, referred to as gauss hybrid models, i.e. GMM.The training sample be divided into skin and
The distribution of two class of non-skin, the training sample is described using the probability density function of two Gauss models:
P (z)=a1g(z；μ1,∑1)+a2g(z；μ2,∑2)；
The parameter of the probability density function is (a1,a2,μ1,μ2,Σ1,Σ2), wherein a1、a2Meet the following conditions:
a1+a2=1；
I is unit matrix, then single Gaussian probability-density function may be expressed as:
Probability of happening expression formula are as follows:
According to the best possibility estimation technique (MLE), the step of deriving calculating parameter, is as follows:
Set an initial parameters valueSettingUse K-means mode
Carry out preliminary classification.The central point for calculating clustering, as μ1,μ2Initial value, the variance conduct of two class of skin and non-skinInitial value；
1, β is calculated using θ1(zi),β2(zi), i=1 ..., n
2, new μ is calculatedjValue:
3, new σ is calculatedjValue:
4, new a is calculatedjValue:
5, the difference for calculating new θ and initial θ, presets tolerance value if it is less than certain, stops, otherwise changing into a new round
Generation.Maximum number of iterations used in the present embodiment is 500, and minimum tolerance value is 1e-10.
For each training sample point, βjThe classification of the maximum point of training sample thus.After completing classification, label two
Class is skin or non-skin, the training sample as next step.
E, the labeled training sample is trained by Adboost algorithm, generates classifier.
In the present embodiment, it will be trained by the training sample of label by Adboost algorithm, by multiple weak typings
Device is combined into strong classifier.Algorithm idea is that multiple training samples generate multiple Weak Classifiers.M iteration, the m times circulation,
Weak Classifier f is trained to according to the sample set of Weighted Coefficientsm(x), the error rate err of current class device is calculatedm, according to error rate
To change classifier weight cmWith the weight w of training samplei。errmIt is bigger, then cmIt is smaller, to increase outstanding Weak Classifier
Weight；And errmIt is bigger, wiAlso bigger, so that lower subseries will focus on the training sample of misclassification class, ultimately generate
The strong classifier being made of M Weak Classifier.
Detailed process is as follows:
Step 1: obtaining labeled training sample (xi,yi), i=1 ..., N, wherein xi∈R,yi∈ -1 ,+1, R are indicated
Training sample, -1 ,+1 indicates skin and two class of non-skin, and N indicates the number of training sample；
Step 2: setting the initial weight w of each training samplei=1/N, i=1 ..., N:
Step 3: circulation executes following steps a, b, c, d, the number of m=1 ..., M, M presentation class device；
Step a. using weights are wiTraining sample training classifier fm(x)∈-1,+1；
Step b. calculates the current erroneous rate of the training sample and the classifierIf errmFor
0, then enable errm=2-52, and according to the error rate errmModify the weight c of the classifierm=log ((1-errm)/errm)；
Step c. is according to the weight c of the classifiermThe weight w that resets the weight of training sample and will resetiIt is normalized, so that
Step d. is using having m classifier calculated gross errors rate, if gross errors rate is 0, end operation；It is no
Then, return step a circulation executes, and circulation executes M end operation when gross errors rate is not 0；
Step 4: obtaining final classifier (i.e. strong classifier)
Further, after image block carries out skin-texture detection in described image to be detected, the present embodiment further include:
Show the image block in described image to be detected with dermatoglyph feature, testing result is as shown in Fig. 2, wherein scheme
2a is the image before detection, and Fig. 2 b is the image after detection.It can be observed from fig. 2 that the present invention for similar in skin color
Clothes, hair, ornaments etc. all have good recognition effect, accuracy rate with higher for the detection of dermatoglyph.
Embodiment two:
Fig. 3 shows the composition knot of the skin-texture detection device provided by Embodiment 2 of the present invention based on Gabor characteristic
Structure, for ease of description, only parts related to embodiments of the present invention are shown.
The skin-texture detection device based on Gabor characteristic can be operate in software unit in terminal device, hard
The part unit unit that perhaps software and hardware combines can also be used as independent pendant and be integrated into the terminal device or run
In the application system of the terminal device.
The skin-texture detection device based on Gabor characteristic specifically includes:
First image block division unit 31, for image to be detected to be divided into multiple images block；
First computing unit 32, for filtering each color channel of each image block with the Gabor pre-established
Wave device group successively does convolution in airspace, obtains k*s*r matrixes identical with the image block size size, wherein k is described
The number in the direction of Gabor filter group, s are the number of the scale of the Gabor filter group, and r is the color channel
Number, k, s, r are the integer greater than zero；
Second computing unit 33, for calculating the mean value and variance of each matrix, and by the mean value and variance
As characteristic value, k*s*r*2 characteristic value is obtained；
Detection unit 34, for using the k*s*r*2 characteristic value as the feature vector of the image block, and by the spy
Sign vector passes through the classifier pre-established and is detected, to determine whether the image block has dermatoglyph feature.
Further, described device further include:
Display unit 35, for showing the image block in described image to be detected with dermatoglyph feature.
Further, described device further include:
Image acquisition units 36, for acquiring multiple training images, the training image includes dermatoglyph feature；
Second image block division unit 37, for the training image to be divided into multiple images block；
Feature extraction unit 38, for obtaining the feature vector of each image block, described eigenvector includes multiple features
Value；
Cluster cell 39, for being clustered, being obtained to described image block using gauss hybrid models according to the characteristic value
Training sample is obtained, and the training sample is labeled as two class of skin and non-skin；
Classifier generation unit 310, for the labeled training sample to be trained by Adboost algorithm,
Generate classifier.
Further, the classifier generation unit 310 includes:
Sample acquisition module 3101, for obtaining labeled training sample (xi,yi), i=1 ..., N, wherein xi∈R,
yi∈ -1 ,+1, R indicate training sample, and -1 ,+1 indicates skin and two class of non-skin, and N indicates the number of training sample；
Weight setup module 3102, for setting the initial weight w of each training samplei=1/N, i=1 ..., N:
Control module 3103 resets submodule for controlled training submodule 31031, computational submodule 31032, weight
31033 and terminate submodule 31034 execution, and terminate submodule 31034 gross errors rate be not 0 when, control institute
State control training submodule 31031, computational submodule 31032, weight reset submodule 31033 and terminate submodule 31034 follows
Ring executes M times, wherein m=1 ..., M, the number of M presentation class device；
Training submodule 31031 is w for using weightsiTraining sample training classifier fm(x)∈-1,+1；
Computational submodule 31032, for calculating the current erroneous rate of the training sample and the classifierIf errmIt is 0, then enables errm=2-52, and according to the error rate errmModify the power of the classifier
Value cm=log ((1-errm)/errm)；
Weight resets submodule 31033, for the weight c according to the classifiermReset the weight of training sampleAnd the weight w that will be resetiIt is normalized, so that
Terminate submodule 31034, has m classifier calculated gross errors rate for utilizing, if gross errors rate is 0,
Then end operation；
Classifier obtains module 3104, for obtaining final classifier
Further, the gauss hybrid models are as follows:
P (z)=a1g(z；μ1,∑1)+a2g(z；μ2,∑2)
Wherein, g (z；μ1,∑1) and g (z；μ2,∑2) indicate that the probability density function of Gauss model, z indicate the feature
Value, a1、a2The coefficient of the probability density function is respectively indicated, wherein a1+a2=1, μ1、μ2Respectively indicate the probability density letter
Several central points, ∑1、∑2The total Variation Matrix of the probability density function is respectively indicated,Indicate variance, I
Expression unit matrix, j=1,2.
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit or module are completed, i.e., the internal structure of described device is divided into different functional unit or module, with complete with
The all or part of function of upper description.Each functional unit in embodiment can integrate in one processing unit, can also be with
It is that each unit physically exists alone, can also be integrated in one unit with two or more units, above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.In addition, each functional unit,
The specific name of module is also only for convenience of distinguishing each other, the protection scope being not intended to limit this application.In above-mentioned apparatus
The specific work process of unit, module, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In conclusion the embodiment of the present invention is by being divided into multiple images block for image to be detected, by each image block
Each color channel and the Gabor filter group of foundation carry out wavelet transformation, and transformation results are extracted into mean value and variance
As characteristic value, so that reducing the Spatial Dimension of feature vector while improving dermatoglyph judging nicety rate, save
Memory headroom.Since the transformation results are matrixes identical with tile size, if by the transformation results directly as
Characteristic value, the Spatial Dimension of feature vector can be very big, thus the embodiment of the present invention by the mean value of matrix and variance as characteristic value,
Spatial Dimension is greatly reduced while retaining main feature.Moreover, the embodiment of the present invention is additionally contemplates that the color ratio of skin
More single, the component on RGB different color channel extracts respectively, so that the extraction accuracy rate than gray level image is higher by very
It is more.In addition, the embodiment of the present invention can choose the Gabor filter group of multiple directions and multiple scales, it is allowed to and image block
Direction and scale are easier to coincide, so that energy is larger after wavelet transformation, better effect can be obtained in skin texture analysis.
Skin-texture detection of embodiment of the present invention accuracy rate is high, speed is fast, it is easy to accomplish, and during entire skin-texture detection not
It needs to increase additional hardware, so as to effectively reduce cost, there is stronger usability and practicality.
In embodiment provided by the present invention, it should be understood that disclosed device and method can pass through others
Mode is realized.For example, the apparatus embodiments described above are merely exemplary, for example, the division of the module or unit,
Only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can be with
In conjunction with or be desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or discussed
Mutual coupling or direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING of device or unit or
Communication connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, the technical solution of the embodiment of the present invention
Substantially all or part of the part that contributes to existing technology or the technical solution can be with software product in other words
Form embody, which is stored in a storage medium, including some instructions use so that one
Computer equipment (can be personal computer, server or the network equipment etc.) or processor (processor) execute this hair
The all or part of the steps of bright each embodiment the method for embodiment.And storage medium above-mentioned include: USB flash disk, mobile hard disk,
Read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic
The various media that can store program code such as dish or CD.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features；And these are modified
Or replacement, the spirit and model of each embodiment technical solution of the embodiment of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (6)
1. a kind of skin-texture detection method based on Gabor characteristic, which is characterized in that the described method includes:
Image to be detected is divided into multiple images block；
Each color channel of each image block is successively done into convolution in airspace with the Gabor filter group pre-established,
K*s*r matrixes identical with the image block size size are obtained, wherein k is the number in the direction of the Gabor filter group,
S is the number of the scale of the Gabor filter group, and r is the number of the color channel, and k, s, r are the integer greater than zero；
The mean value and variance of each matrix are calculated, and using the mean value and variance as characteristic value, obtains k*s*r*2
Characteristic value；
Using the k*s*r*2 characteristic value as the feature vector of the image block, and by described eigenvector by pre-establishing
Classifier detected, with determine the image block whether there is dermatoglyph feature；
The classifier of establishing includes:
Multiple training images are acquired, the training image includes dermatoglyph feature；
The training image is divided into multiple images block；
The feature vector of each image block is obtained, described eigenvector includes multiple characteristic values；
According to the characteristic value, described image block is clustered using gauss hybrid models, obtains training sample, and will be described
Training sample is labeled as two class of skin and non-skin；
The labeled training sample is trained by Adboost algorithm, generates classifier；
The labeled training sample is trained by Adboost algorithm, generating classifier includes:
Step 1: obtaining labeled training sample (xi,yi), i=1 ..., N, wherein xi∈R,yi∈ -1 ,+1, R indicate training
Sample, -1 ,+1 indicates skin and two class of non-skin, and N indicates the number of training sample；
Step 2: setting the initial weight w of each training samplei=1/N, i=1 ..., N:
Step 3: circulation executes following steps a, b, c, d, the number of m=1 ..., M, M presentation class device；
Step a. using weights are wiTraining sample training classifier fm(x)∈-1,+1；
Step b. calculates the current erroneous rate of the training sample and the classifierIf errmIt is 0, then
Enable errm=2-52, and according to the error rate errmModify the weight c of the classifierm=log ((1-errm)/errm)；
Step c. is according to the weight c of the classifiermReset the weight of training sample
And the weight w that will be resetiIt is normalized, so that
Step d. is using having m classifier calculated gross errors rate, if gross errors rate is 0, end operation；Otherwise, it returns
It returns step a circulation to execute, and circulation executes M end operation when gross errors rate is not 0；
Step 4: obtaining final classifier
2. the method as described in claim 1, which is characterized in that the method also includes:
Show the image block in described image to be detected with dermatoglyph feature.
3. the method as described in claim 1, which is characterized in that the gauss hybrid models are as follows:
P (z)=a1g(z；μ1,∑1)+a2g(z；μ2,∑2)
Wherein, g (z；μ1,∑1) and g (z；μ2,∑2) indicate that the probability density function of Gauss model, z indicate the characteristic value, a1、
a2The coefficient of the probability density function is respectively indicated, wherein a1+a2=1, μ1、μ2It respectively indicates in the probability density function
Heart point, ∑1、∑2The total Variation Matrix of the probability density function is respectively indicated, Indicate variance, I indicates unit
Matrix, j=1,2.
4. a kind of skin-texture detection device based on Gabor characteristic, which is characterized in that described device includes:
First image block division unit, for image to be detected to be divided into multiple images block；
First computing unit, for by each color channel of each image block and the Gabor filter group that pre-establishes
Convolution successively is done in airspace, obtains k*s*r matrixes identical with the image block size size, wherein k is Gabor filtering
The number in the direction of device group, s are the number of the scale of the Gabor filter group, and r is the number of the color channel, k, s, r
For the integer greater than zero；
Second computing unit, for calculating the mean value and variance of each matrix, and using the mean value and variance as spy
Value indicative obtains k*s*r*2 characteristic value；
Detection unit, for using the k*s*r*2 characteristic value as the feature vector of the image block, and by described eigenvector
It is detected by the classifier pre-established, to determine whether the image block has dermatoglyph feature；
Image acquisition units, for acquiring multiple training images, the training image includes dermatoglyph feature；
Second image block division unit, for the training image to be divided into multiple images block；
Feature extraction unit, for obtaining the feature vector of each image block, described eigenvector includes multiple characteristic values；
Cluster cell, for being clustered, being trained to described image block using gauss hybrid models according to the characteristic value
Sample, and the training sample is labeled as two class of skin and non-skin；
Classifier generation unit generates classification for the labeled training sample to be trained by Adboost algorithm
Device；
The classifier generation unit includes:
Sample acquisition module, for obtaining labeled training sample (xi,yi), i=1 ..., N, wherein xi∈R,yi∈-1,+
1, R indicates training sample, and -1 ,+1 indicates skin and two class of non-skin, and N indicates the number of training sample；
Weight setup module, for setting the initial weight w of each training samplei=1/N, i=1 ..., N:
Control module resets submodule for controlled training submodule, computational submodule, weight and terminates holding for submodule
Row, and when terminating the gross errors rate of submodule is not 0, control the controlled training submodule, computational submodule, weight weight
If submodule and terminating submodule circulation and executing M times, wherein m=1 ..., M, the number of M presentation class device；
Training submodule is w for using weightsiTraining sample training classifier fm(x)∈-1,+1；
Computational submodule, for calculating the current erroneous rate of the training sample and the classifierIf
errmIt is 0, then enables errm=2-52, and according to the error rate errmModify the weight c of the classifierm=log ((1-errm)/
errm)；
Weight resets submodule, for the weight c according to the classifiermReset the weight of training sampleAnd the weight w that will be resetiIt is normalized, so that
Terminate submodule, for if gross errors rate is 0, terminating to grasp using having m classifier calculated gross errors rate
Make；
Classifier obtains module, for obtaining final classifier
5. device as claimed in claim 4, which is characterized in that described device further include:
Display unit, for showing the image block in described image to be detected with dermatoglyph feature.
6. device as claimed in claim 4, which is characterized in that the gauss hybrid models are as follows:
P (z)=a1g(z；μ1,∑1)+a2g(z；μ2,∑2)
Wherein, g (z；μ1,∑1) and g (z；μ2,∑2) indicate that the probability density function of Gauss model, z indicate the characteristic value, a1、
a2The coefficient of the probability density function is respectively indicated, wherein a1+a2=1, μ1、μ2It respectively indicates in the probability density function
Heart point, ∑1、∑2The total Variation Matrix of the probability density function is respectively indicated, Indicate variance, I indicates unit
Matrix, j=1,2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410203716.1A CN105095836B (en) | 2014-05-14 | 2014-05-14 | A kind of skin-texture detection method and device based on Gabor characteristic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410203716.1A CN105095836B (en) | 2014-05-14 | 2014-05-14 | A kind of skin-texture detection method and device based on Gabor characteristic |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105095836A CN105095836A (en) | 2015-11-25 |
| CN105095836B true CN105095836B (en) | 2019-03-01 |
Family
ID=54576230
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410203716.1A Active CN105095836B (en) | 2014-05-14 | 2014-05-14 | A kind of skin-texture detection method and device based on Gabor characteristic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105095836B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105869171A (en) * | 2016-04-18 | 2016-08-17 | 重庆大学 | Method for quantitatively analyzing human skin line aging |
| CN108272533B (en) * | 2017-12-26 | 2019-12-17 | 中国科学院苏州生物医学工程技术研究所 | Skin modeling method for skin wound area |
| CN108036746B (en) * | 2017-12-26 | 2019-08-06 | 太原理工大学 | A kind of Gabor transformation realization carbon fibre composite surface texture analysis method based on Spectrum Method |
| CN109308458B (en) * | 2018-08-31 | 2022-03-15 | 电子科技大学 | Method for improving small target detection precision based on characteristic spectrum scale transformation |
| CN109829434A (en) * | 2019-01-31 | 2019-05-31 | 杭州创匠信息科技有限公司 | Method for anti-counterfeit and device based on living body texture |
| CN110232404A (en) * | 2019-05-21 | 2019-09-13 | 江苏理工学院 | A kind of recognition methods of industrial products surface blemish and device based on machine learning |
| CN112149468A (en) * | 2019-06-28 | 2020-12-29 | 瑞昱半导体股份有限公司 | Color gamut weight detection method and device for skin color image |
| CN110569873A (en) * | 2019-08-02 | 2019-12-13 | 平安科技（深圳）有限公司 | Image recognition model training method and device and computer equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509097A (en) * | 2011-09-29 | 2012-06-20 | 北京新媒传信科技有限公司 | Method and device for image segmentation |
| CN103294983A (en) * | 2012-02-24 | 2013-09-11 | 北京明日时尚信息技术有限公司 | Scene recognition method in static picture based on partitioning block Gabor characteristics |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7466846B2 (en) * | 2002-09-25 | 2008-12-16 | The Hong Kong Polytechnic University | Method for analyzing a palm print for the identification of an individual using gabor analysis |
-
2014
- 2014-05-14 CN CN201410203716.1A patent/CN105095836B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102509097A (en) * | 2011-09-29 | 2012-06-20 | 北京新媒传信科技有限公司 | Method and device for image segmentation |
| CN103294983A (en) * | 2012-02-24 | 2013-09-11 | 北京明日时尚信息技术有限公司 | Scene recognition method in static picture based on partitioning block Gabor characteristics |
Non-Patent Citations (3)
| Title |
|---|
| 一种新的基于聚类的多分类器融合算法;刘汝杰 等;《计算机研究与发展》;20011030;第38卷(第10期);第1236-1241页 * |
| 一种采用Gabor小波的纹理特征提取方法;张刚等;《中国图象图形学报》;20100228;第15卷(第2期);第247-254页 * |
| 基于AdaBoost的组合分类器在遥感影像分类中的应用;周红英 等;《计算机应用研究》;20071015;第24卷(第10期);第182-185页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105095836A (en) | 2015-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105095836B (en) | A kind of skin-texture detection method and device based on Gabor characteristic | |
| CN107564025B (en) | Electric power equipment infrared image semantic segmentation method based on deep neural network | |
| Yuan et al. | Factorization-based texture segmentation | |
| CN107979554B (en) | Radio signal Modulation Identification method based on multiple dimensioned convolutional neural networks | |
| CN109117826B (en) | Multi-feature fusion vehicle identification method | |
| CN110244271B (en) | Radar radiation source sorting and identifying method and device based on multiple synchronous compression transformation | |
| CN105574534B (en) | Conspicuousness object detection method based on sparse subspace clustering and low-rank representation | |
| CN110163813B (en) | Image rain removing method and device, readable storage medium and terminal equipment | |
| CN103208097B (en) | Filtering method is worked in coordination with in the principal component analysis of the multi-direction morphosis grouping of image | |
| CN104268593A (en) | Multiple-sparse-representation face recognition method for solving small sample size problem | |
| CN111127387B (en) | Quality evaluation method for reference-free image | |
| CN107301643B (en) | Well-marked target detection method based on robust rarefaction representation Yu Laplce's regular terms | |
| CN112287784B (en) | Radar signal classification method based on deep convolutional neural network and feature fusion | |
| CN108664911A (en) | A kind of robust human face recognition methods indicated based on image sparse | |
| CN113095333B (en) | Unsupervised feature point detection method and unsupervised feature point detection device | |
| CN114663685B (en) | Pedestrian re-recognition model training method, device and equipment | |
| CN107067407B (en) | Contour detection method based on non-classical receptive field and linear nonlinear modulation | |
| CN104156628A (en) | Ship radiation signal recognition method based on multi-kernel learning and discriminant analysis | |
| CN109829412A (en) | The Partial Discharge Pattern Recognition Method of fractal characteristic is decomposed based on dynamic mode | |
| CN110796022A (en) | Low-resolution face recognition method based on multi-manifold coupling mapping | |
| Niu et al. | Siamese-network-based learning to rank for no-reference 2D and 3D image quality assessment | |
| CN114429151A (en) | Magnetotelluric signal identification and reconstruction method and system based on depth residual error network | |
| CN105631441A (en) | Human face recognition method | |
| CN113673465A (en) | Image detection method, device, equipment and readable storage medium | |
| CN105389573B (en) | A kind of face identification method based on three value mode layering manufactures of part |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |