CN105005756A - Method and device for face identification - Google Patents

Abstract

The invention relates to the field of security and protection, and especially to a method and device for face identification. The method and device can improve whole properties of a face identification system in the prior art. The method comprises: performing discrete cosine transform (DCT) of a to-be-indentified face image to obtain a feature vector of a frequency domain image; performing ILDP operation of the to-be-indentified face image to obtain a feature vector of a space domain image; fusing the feature vector of the frequency domain image and the feature vector of the space domain image; and performing face identification of the fused feature vector. According to the embodiment of the invention, the method and device for face identification protect relative information of space domain and frequency domain through fusing DCT and ILDP technology, and the properties of the face identification system are improved by utilizing mutual complementation of space domain information and frequency domain information.

Description

A kind of face identification method and device

Technical field

The present invention relates to safety-security area, particularly relate 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.Image or the video flowing of face is contained with video camera or camera collection, and automatic detection and tracking face in the picture, and then the face detected is carried out to a series of correlation techniques of face, be usually also called Identification of Images, face recognition.

The mode of carrying out recognition of face at present has a variety of, comprising:

One, in face recognition technology, be used alone discrete cosine transform (DCT) method operate.DCT is a kind of conventional orthogonal transformation, and it is transformed from a kind of Fourier transform of special shape, when original signal meets some requirements, just Fourier transform can be converted into cosine transform, and its transformation kernel is cosine function;

Two, in face recognition technology, use local orientation's pattern (LDP).LDP is a kind of texture description operator, carries out calculating LDP code value by the skirt response value calculating eight directions, obtains LDP histogram as feature to be identified by LDP code value.

If the present inventor finds to be used alone the loss that DCT method will cause spatial information (si); And use LDP pattern also can there is the insufficient problem of sampling, that does not also consider graded moves towards problem simultaneously.Therefore the system performance of face recognition technology of the prior art needs to improve further.

Summary of the invention

A kind of face identification method that the embodiment of the present invention provides and device thereof, for improving the overall performance of face identification system in prior art.

A kind of face identification method, described method comprises:

Discrete cosine transform operation is carried out to the facial image of pending identification, obtains the proper vector of frequency domain figure picture;

ILDP operation is carried out to the facial image of described pending identification, obtains the proper vector of spatial domain picture;

The proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture are merged; And recognition of face is carried out to the proper vector after merging.

A kind of face identification device, described device comprises:

Frequency domain module, for carrying out discrete cosine transform operation to the facial image of pending identification, obtains the proper vector of frequency domain figure picture;

Spatial domain module, for carrying out ILDP operation to the facial image of described pending identification, obtains the proper vector of spatial domain picture;

Fusion Module, for merging the proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture; And recognition of face is carried out to the proper vector after merging.

The face identification method that the visible embodiment of the present invention provides and device thereof, combine DCT and ILDP technology and can save the relevant information in spatial domain and frequency domain from damage, and the characteristic simultaneously utilizing spatial information (si) and frequency domain information mutually to supplement improves the performance of face identification system.

Term " first ", " second ", " the 3rd " " 4th " etc. (if existence) in instructions of the present invention and claims and above-mentioned accompanying drawing are for distinguishing similar object, and need not be used for describing specific order or precedence.The embodiments described herein should be appreciated that the data used like this can be exchanged in the appropriate case, so that can be implemented with the order except the content except here diagram or description.In addition, term " comprises " and " having " and their any distortion, intention is to cover not exclusive comprising, such as, contain those steps or unit that the process of series of steps or unit, method, system, product or equipment is not necessarily limited to clearly list, but can comprise clearly do not list or for intrinsic other step of these processes, method, product or equipment or unit.

The above, above embodiment only in order to technical scheme of the present invention to be described, is not intended to limit; Although with reference to previous embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein portion of techniques feature; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.

Accompanying drawing explanation

The recognition of face process flow diagram that Fig. 1 provides for the embodiment of the present invention;

The facial image of pending identification of Fig. 2 (a) for using in the embodiment of the present invention;

Fig. 2 (b) is the facial image in the embodiment of the present invention after DCT conversion;

Fig. 3 is the well type Neighborhood Graph mentioned in the embodiment of the present invention;

Fig. 4 is the complete computation process flow diagram flow chart of the ILDP code mentioned in the embodiment of the present invention;

The structural drawing of a kind of face identification device that Fig. 5 provides for the embodiment of the present invention.

Embodiment

Embodiments provide a kind of invasion and boundary defence method, entire image processing time longer poor real and the generous tropism of rate of false alarm in prior art can be solved and determine more difficult problem.

The present invention program is understood better in order to make those skilled in the art person, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the embodiment of a part of the present invention, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, should belong to the scope of protection of the invention.

The face identification method that the embodiment of the present invention provides and device thereof, combine DCT and ILDP technology and can save the relevant information in spatial domain and frequency domain from damage, and the characteristic simultaneously utilizing spatial information (si) and frequency domain information mutually to supplement improves the performance of face identification system.As shown in Figure 1, detailed process is as follows:

Step 11, carries out discrete cosine transform operation to the facial image of pending identification, obtains the proper vector of frequency domain figure picture;

Step 12, carries out ILDP operation to the facial image of described pending identification, obtains the proper vector of spatial domain picture;

Step 13, merges the proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture; And recognition of face is carried out to the proper vector after merging.

Concrete, the described facial image to pending identification carries out discrete cosine transform operation, and the proper vector obtaining frequency domain figure picture comprises:

The facial image of pending identification is carried out discrete cosine transform operation, and in the scope that low frequency coefficient is assembled, dimensionality reduction operation is carried out in selected part region, obtains the proper vector of frequency domain figure picture.

Concrete, the described facial image to described pending identification carries out ILDP operation, and the proper vector obtaining spatial domain picture comprises:

Be ILDP code value according to presetting method by each pixel transitions of the facial image of described pending identification, obtain the proper vector of spatial domain picture according to described ILDP code value.

Concrete, described is that ILDP code value comprises according to presetting method by each pixel transitions of the facial image of described pending identification:

Centered by each pixel of the facial image of pending identification, obtain the well type neighborhood of this pixel, be converted to the ILDP code value of this pixel according to this well type neighborhood.

Concrete, the described proper vector obtaining spatial domain picture according to described ILDP code value comprises:

Judge the number of times that each ILDP code value occurs, and using the proper vector of all number of times as spatial domain picture.

Be described with specific embodiment below:

Embodiment:

The embodiment of the present invention provides a kind of face identification method, DCT and ILDP technology can effectively combine by the method, and utilize the characteristic of spatial domain and frequency domain complementation to improve the performance of face identification system, detailed process comprises:

Steps A, carries out discrete cosine transform operation to the facial image of pending identification, obtains the proper vector of frequency domain figure picture; Both the facial image of pending identification was carried out discrete cosine transform operation, in the scope that low frequency coefficient is assembled, dimensionality reduction operation is carried out in selected part region, obtains the proper vector of frequency domain figure picture.

This concrete step process is as follows:

The main thought of this step is: DCT is a kind of popular transformation tool.Data transformation to frequency domain, then as required, is ignored a part of high-frequency information, is retained main low-frequency information, thus data volume is reduced by DCT.Utilize this characteristic significantly can reduce the calculated amount of data;

The facial image of pending identification is after DCT conversion, and main information concentrates in minority low frequency coefficient, therefore can ignore some high frequency coefficients according to this characteristic, only selects a low frequency DCT coefficients subset as the feature of facial image.As shown in Figure 2, wherein a is the facial image of pending identification, and b is the facial image after DCT conversion.As can be seen from Figure, facial image is after dct transform, and the coefficient that numerical value is large concentrates on the low frequency part in the upper left corner, and the main information amount namely after dct transform concentrates on less low frequency coefficient.Consider that computing time is less, the method still adopting typical DCT coefficient to choose herein, the little square sub blocks choosing of the image upper left corner as feature to carry out dimensionality reduction.Need the correlation computations of carrying out as follows in this step:

A given length is list entries u (n) of N.Shown in its following formula (1) of discrete cosine transform expression formula v (k):

$v\left(k\right)=a\left(k\right)\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}u\left(n\right)\cos \left[\frac{(2n+1)\mathrm{\πk}}{2N}\right]---\left(1\right)$

Shown in following formula (2), wherein: (0≤k≤N 1).

$\left\{\begin{array}{c}a\left(0\right)=\sqrt{1/N},\\ a\left(k\right)=\sqrt{2/N},\end{array}\right.(1\≤k\≤N-1)---\left(2\right)$

Sequence u (n) is regarded as vector, and DCT spatial domain Matrix C={ c (k, n) } is expressed as shown in formula (3):

$\left\{\begin{array}{c}1/\sqrt{N}\\ (k=\mathrm{0,0}\≤n\≤N-1)\\ \sqrt{2/N}\cos \left[\frac{(2n+1)\mathrm{\πk}}{2N}\right]\\ (1\≤k\≤N-\mathrm{1,0}\≤n\≤N-1)\end{array}\right.---\left(3\right)$

Wherein, k and n is that number and row ordinal number respectively, and the DCT of such sequence u (n) is reduced to following formula (4):

v＝Cu. (4)

By discrete cosine transform, a sequence is broken down into the weighted sum of cosine basis sequence, and these cosine basis sequences are exactly the row vector of Matrix C.Matrix C is a collection matrix, so DCT is orthogonal transformation.Therefore facial image obtains one with original image DCT proper vector of a size coefficient image after DCT, wherein the proper vector of DCT proper vector both frequency domain figure picture.

Step B, carries out ILDP operation to the facial image of described pending identification, obtains the proper vector of spatial domain picture; Both be ILDP code value according to presetting method by each pixel transitions of the facial image of described pending identification, obtained the proper vector of spatial domain picture according to described ILDP code value.

Wherein, be ILDP code value according to presetting method by each pixel transitions of the facial image of described pending identification, refer to: centered by each pixel of the facial image of pending identification, obtain the well type neighborhood of this pixel, be converted to the ILDP code value of this pixel according to this well type neighborhood, in this step, detailed process is as follows:

For investigating the wider gradient information in local, devise local well type neighborhood, if the well type neighborhood of facial image central pixel point (i, j) is see Fig. 3.Well type neighborhood sample range compared with 3 × 3 traditional neighborhoods is wider, comprises sample radius R=1 and R=2; And sampled point is less compared with 5 × 5 traditional neighborhoods, only have 16 pixels.Meanwhile, centered by (i, j) point, 16 sampled point place radial lines of well type neighborhood are not overlapping, which ensure that the redundancy avoiding information under the sufficient prerequisite of sampling.

Well type neighborhood 16 pixels of local center pixel are divided into two groups of { P according to sample radius (R=1 or 2) by ILDP _iand { Q _i(i=1 ..., 7) and be arranged in two 3 × 3 subneighborhoods, by { P _iand { Q _irespectively with 8 Kirsch mask convolutions, obtain two groups of edge gradient value { p _iand { q _i.Edge gradient value p _iand q _isign represent two relatively independent graded trend, positive sign represents that gradient rises, and negative sign represents Gradient Descent.Add facial image local pixel gray-scale value and there is correlativity in spatial domain, therefore ILDP not edge ask absolute value, and directly utilize { p _iand { q _iin the direction of respective maximal value as local feature information, use p simultaneously _ior q _isubscript i represent p _ior q _ithe direction at place.If p _a{ p _imaximal value, q _b{ q _imaximal value, the ILDP pattern of center pixel can with the octal numeral (ab) of two ₈represent, claim this number to be ILDP code, decimal number can be converted to by formula (5).

ILDPcode＝a×8 ¹+b×8 ⁰(5)

Wherein all desirable 8 values of a and b, all ILDP patterns have 8 ²=64 kinds.Fig. 4 describes the complete computation process of ILDP code.All pixels of a facial image obtain well type neighborhood union respectively and convert ILDP code to centered by oneself, and the ILDP code of all pixels forms corresponding ILDP figure according to former figure coordinate arrangement, and this facial image is with regard to the ILDP figure representative of available correspondence.

The proper vector process obtaining spatial domain picture according to described ILDP code value is as follows:

ILDP histogram feature can be obtained by ILDP code value, both judge the number of times that each ILDP code value occurs, and using the proper vector of all number of times as spatial domain picture.The histogram calculation formula (6) of corresponding ILDP is as follows:

$H_{{\mathrm{ILDP}}_I}=\underset{X,Y}{\mathrm{\Σ}}f({\mathrm{ildp}}_k(x,y),c_i)---\left(6\right)$

Wherein c _iit is ILDP code value.

Step C, merges the proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture; And recognition of face is carried out to the proper vector after merging.This step detailed process is as follows:

DCT proper vector (both the proper vector of frequency domain figure picture) and ILDP proper vector (both the proper vector of spatial domain picture) merge.Assuming that DCT proper vector after dimensionality reduction and ILDP proper vector are y respectively ₁∈ R _m1and y ₂∈ R _m2, then the assemblage characteristic vector z ∈ R after normalizing _m1+m2be expressed as shown in formula (7):

z＝[y ₁/σ ₁y ₁/σ ₂] ^T. (7)

σ in formula ₁and σ ₂y respectively ₁and y ₂standard deviation, can calculate by asking the square root of the variance of proper vector.

Finally carry out recognition of face by nearest neighbor method sorting technique.

As shown in Figure 5, the embodiment of the present invention provides a kind of face identification device, comprising:

Frequency domain module 51, for carrying out discrete cosine transform operation to the facial image of pending identification, obtains the proper vector of frequency domain figure picture;

Spatial domain module 52, for carrying out ILDP operation to the facial image of described pending identification, obtains the proper vector of spatial domain picture;

Fusion Module 53, for merging the proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture; And recognition of face is carried out to the proper vector after merging.

Described frequency domain module 51 specifically for:

The facial image of pending identification is carried out discrete cosine transform operation, and in the scope that low frequency coefficient is assembled, dimensionality reduction operation is carried out in selected part region, obtains the proper vector of frequency domain figure picture.

Described spatial domain module 52 specifically for:

Be ILDP code value according to presetting method by each pixel transitions of the facial image of described pending identification, obtain the proper vector of spatial domain picture according to described ILDP code value.

Described spatial domain module 52 specifically for:

Centered by each pixel of the facial image of pending identification, obtain the well type neighborhood of this pixel, be converted to the ILDP code value of this pixel according to this well type neighborhood.

Described spatial domain module 52 specifically for:

Judge the number of times that each ILDP code value occurs, and using the proper vector of all number of times as spatial domain picture.

Local orientation's pattern (LDP) involved herein: LDP is a kind of texture description operator, carries out calculating LDP code value by the skirt response value calculating eight directions, obtains LDP histogram as feature to be identified by LDP code value.

The local orientation's pattern (ILDP) improved: be retain LDP one on advantageous basis improve one's methods, be to all pixels of piece image all centered by oneself pixel obtain well type neighborhood union and change into ILDP code, the ILDP code of all pixels forms ILDP figure according to former figure coordinate arrangement, this facial image just ILDP of available correspondence schemes to represent, therefore obtains ILDP histogram as feature to be identified by ILDP code value.

Kirsch operator: the directional operator be made up of 8 templates representing 8 directions is also the process considering image 3 × 3 neighborhood.Make these 8 templates of each pixel in image carry out convolutional calculation, wherein maximal value can be used as the output of edge image.

Joint histogram: CBP histogram and LDP histogram are linked in sequence the total characteristic that formation will identify.

In sum, beneficial effect:

A kind of face identification method merging local orientation's pattern (ILDP) feature of face discrete cosine transform (DCT) and improvement is proposed.DCT is transformed into frequency domain facial image, and the main information of facial image concentrates in minority DCT coefficient, and choosing a low frequency DCT coefficients subset containing face main information can as the frequency domain character of face.ILDP is the operator of a fine-scale, Description Image regional area texture variations, by carrying out the ILDP coding of Pixel-level to facial image, then asks for the statistic histogram of ILDP coding, can obtain the ILDP face characteristic with space invariance.DCT and ILDP has the complementarity between frequency domain and spatial domain; The grain details of the ILDP feature interpretation of image often shows as high-frequency information, and this and low frequency DCT coefficients also have complementarity.DCT and ILDP is efficient face characteristic describing method, two kinds of performances having complementary Fusion Features and be conducive to improving face identification system.

Have above-mentioned car of crossing also can find out, the present invention because the numeric ratio that will calculate is less, therefore significantly can reduce calculated amount in computation process.The present invention has simultaneously taken into full account the adequacy of sampling and has avoided information redundancy.Good robustness is had to noise.Utilize gradient information relatively stable, and consider the trend trend of graded.

Each embodiment in this instructions all adopts the mode of going forward one by one to describe, between each embodiment identical similar part mutually see, what each embodiment stressed is the difference with other embodiments.Especially, for device embodiment, describe fairly simple, relevant part illustrates see the part of embodiment of the method.The above is only the specific embodiment of the present invention, and for those skilled in the art, under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (10)

1. a face identification method, is characterized in that, described method comprises:

Discrete cosine transform operation is carried out to the facial image of pending identification, obtains the proper vector of frequency domain figure picture;

ILDP operation is carried out to the facial image of described pending identification, obtains the proper vector of spatial domain picture;

The proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture are merged; And recognition of face is carried out to the proper vector after merging.

2. the method for claim 1, is characterized in that, the described facial image to pending identification carries out discrete cosine transform operation, and the proper vector obtaining frequency domain figure picture comprises:

The facial image of pending identification is carried out discrete cosine transform operation, and in the scope that low frequency coefficient is assembled, dimensionality reduction operation is carried out in selected part region, obtains the proper vector of frequency domain figure picture.

3. the method for claim 1, is characterized in that, the described facial image to described pending identification carries out ILDP operation, and the proper vector obtaining spatial domain picture comprises:

Be ILDP code value according to presetting method by each pixel transitions of the facial image of described pending identification, obtain the proper vector of spatial domain picture according to described ILDP code value.

4. method as claimed in claim 3, is characterized in that, described is that ILDP code value comprises according to presetting method by each pixel transitions of the facial image of described pending identification:

Centered by each pixel of the facial image of pending identification, obtain the well type neighborhood of this pixel, be converted to the ILDP code value of this pixel according to this well type neighborhood.

5. method as claimed in claim 3, it is characterized in that, the described proper vector obtaining spatial domain picture according to described ILDP code value comprises:

Judge the number of times that each ILDP code value occurs, and using the proper vector of all number of times as spatial domain picture.

6. a face identification device, is characterized in that, described device comprises:

Frequency domain module, for carrying out discrete cosine transform operation to the facial image of pending identification, obtains the proper vector of frequency domain figure picture;

Spatial domain module, for carrying out ILDP operation to the facial image of described pending identification, obtains the proper vector of spatial domain picture;

Fusion Module, for merging the proper vector of described frequency domain figure picture and the proper vector of described spatial domain picture; And recognition of face is carried out to the proper vector after merging.

7. method as claimed in claim 6, is characterized in that, described frequency domain module specifically for:

The facial image of pending identification is carried out discrete cosine transform operation, and in the scope that low frequency coefficient is assembled, dimensionality reduction operation is carried out in selected part region, obtains the proper vector of frequency domain figure picture.

8. method as claimed in claim 7, is characterized in that, described spatial domain module specifically for:

Be ILDP code value according to presetting method by each pixel transitions of the facial image of described pending identification, obtain the proper vector of spatial domain picture according to described ILDP code value.

9. method as claimed in claim 8, is characterized in that, described spatial domain module specifically for:

Centered by each pixel of the facial image of pending identification, obtain the well type neighborhood of this pixel, be converted to the ILDP code value of this pixel according to this well type neighborhood.

10. method as claimed in claim 8, is characterized in that, described spatial domain module specifically for:

Judge the number of times that each ILDP code value occurs, and using the proper vector of all number of times as spatial domain picture.