CN1477588A - Automatic human face identification method based on personal image - Google Patents

Abstract

The present invention discloses an automatic face identification method based on personal image, it includes the utilization of digital image input equipment to obtain digital gray level image, then utilizes the information of eye location to obtain human face region. It also provides the concrete steps of said human face identification method, including training identification mechanism, generating training image, adding projected image, receiving image to be identified, comparing to obtain identificaltion result.

Description

A kind of automatic human face identification method based on everyone piece image

One, technical field

The present invention relates to a kind of digital facial image and detect and recognition device, particularly a kind of each object to be identified that only needs has a width of cloth training image just can carry out the method for Automatic face recognition.

Two, background technology

The numeral facial image detects and recognition device, can be widely used in aspects such as identity document identification, buildings access and exit control, computer log control, credit cardholder's discriminating, criminal's tracking, accident detection, carry out the method for identity discriminating with other human body biological characteristics such as utilizing fingerprint, iris and compare, end user's face is differentiated friendly more and convenient.But in some specific application, for example public security organ pursues and captures an escaped prisoner, each object to be identified often has only piece image available, and present digital facial image recognition technology need be prepared multiple image for each object to be identified and recognition mechanism trained being used for, and this just makes existing digital facial image detect with recognition device to be difficult to handle well this generic task.

Three, summary of the invention

The objective of the invention is needs each object to be identified to have multiple image just can carry out the problem of recognition of face at prior art, provide a kind of each object to be identified only need have piece image just can carry out the method for Automatic face recognition, improve digital facial image and detect performance with recognition device with auxiliary.

For realizing purpose of the present invention, the invention provides a kind of method of utilizing everyone width of cloth digital gray scale training image to carry out Automatic face recognition, this method may further comprise the steps: (1) if recognition mechanism does not train, then execution in step 2, otherwise forward step 4 to; (2) generate the projected image that adds of training image; (3) utilize the principal component analysis (PCA) technology to generate the feature space that adds projected image; (4) receive image to be identified; (5) generate the projected image that adds of image to be identified; (6) projected image that adds that adds projected image and training image for the treatment of recognition image in adding the projected image feature space is compared; (7) with the pairing object of training image of comparison similarity maximum as recognition result; (8) finish.

Advantage of the present invention is only need have piece image just can carry out Automatic face recognition, thus the auxiliary performance that has improved digital facial image detection and recognition device.

Below in conjunction with accompanying drawing most preferred embodiment is elaborated.

Four, description of drawings

Fig. 1 is that digital facial image detects and the recognition device workflow diagram.

Fig. 2 is the process flow diagram of the inventive method.

Fig. 2 generates the process flow diagram that adds projected image.

Five, embodiment

As shown in Figure 1, digital facial image detects with recognition device and obtains gray level image by Digital Image Input Device, utilizes the information of eye location to obtain human face region then.Just handle then, generally include facial image is carried out feature extraction and compares with the existing image in the facial image database by recognition mechanism.

Method of the present invention as shown in Figure 2.Step 10 is initial actuatings.Step 11 judges whether recognition mechanism trains, if execution in step 17 then; Otherwise execution in step 12.Step 12 is obtained a width of cloth training image.Step 13 generates the projected image that adds of this image, and adding projected image is the image that obtains after handling with the custom-designed processing mode of the present invention, and this step will be specifically introduced in conjunction with Fig. 3 in the part of back.

Though each object to be identified has only piece image, object to be identified has a plurality of, so the step 14 of Fig. 2 judges whether in addition other training images, if having then forward step 12 to; Otherwise execution in step 15.It is raw data that step 15 adds projected image with step 12 to 14 one groups of producing, and utilizes the principal component analysis (PCA) technology in the higher algebra, generate K orthogonal basis vector arranged add the projection properties space, K is the round values for example 20 of user preset here.Step 16 projects to all training images respectively and adds in the projection properties space, thereby obtain its each self-corresponding projection properties vector that adds, and preserved, played the effect of facial image database among Fig. 1 on the collective entity that the proper vector of these preservations is formed.

The step 17 of Fig. 2 receives facial image to be identified, and 13 way generates the projected image that adds of this image set by step then.Step 18 with the projector, image projection that adds to be identified to adding in the projection properties space, thereby obtain its correspondence add the projection properties vector.In fact step 17 has played the effect of facial image feature extraction among Fig. 1 to 19 processing.Step 19 will be waited to know the projection properties vector that adds of image and compare with the projection properties vector that adds of each training image.Two vectors are compared and can be finished by the inner product of asking vector, and the more little then vector of inner product is similar more.Step 20 find out to wait to know image add the most similar training image of projection properties vector add the projection properties vector, and with the object (being the owner of people's face) under this training image as recognition result.Step 21 is done states.

Fig. 3 describes the step 13 of Fig. 2 in detail, and its effect is the projected image that adds that generates a width of cloth gray level image.The step 130 of Fig. 3 is initial states.Step 131 is obtained gray level image.Step 132 calculates the vertical projection and the horizontal projection of image.Suppose that P is a width of cloth N ₁* N ₂The gray level image of size, P (x, y) expression P mid point (x, the gray-scale value of y) locating, x ∈ [1, N ₁|, y ∈ [1, N ₂], point (x, the vertical projection V that y) locates then _P(x) and horizontal projection H _P(y) calculate by following formula respectively: $V_P\left(x\right)=\underset{y=1}{\overset{N_2}{\mathrm{\Σ}}}P(x,y)$ $H_P\left(y\right)=\underset{x=1}{\overset{N_1}{\mathrm{\Σ}}}P(x,y)$

The step 133 of Fig. 3 generates the perspective view M of original image (being P) _P, M _PAt point (x, the gray-scale value M that y) locates _P(x, y) calculate by the custom-designed formula of the present invention: $M_P(x,y)=\frac{V_P\left(x\right)H_P\left(y\right)}{N_1{N}_2\stackrel{\‾}{P}}$ Wherein P is the average gray value of P: $\stackrel{\‾}{P}=\frac{\underset{x=1}{\overset{N_1}{\mathrm{\Σ}}}\underset{y=1}{\overset{N_2}{\mathrm{\Σ}}}P(x,y)}{N_1{N}_2}$

What the step 134 of Fig. 3 generated original image adds projected image P _α, P _αAt point (x, the gray-scale value P that y) locates _α(x, y) calculate by the custom-designed formula of the present invention: $P_{\mathrm{\α}}(x,y)=\frac{P(x,y)+\mathrm{\α}{M}_P(x,y)}{1+\mathrm{\α}}$ Wherein α is the fractional value for example 0.25 between (0,1) of user preset.

The step 135 of Fig. 3 is done states of Fig. 3.

Because perspective view M _PMid point (x, the gray-scale value of y) locating are determined by the average gray of the row and column at this place in the former gray level image, therefore, and when α is not too big, with M _PWith P combine generate add projected image P _αIn keeping P, in the main information, P has been carried out obfuscation.When each object to be identified had multiple image, this Fuzzy processing had been abandoned the utilizable information of some possibilities, may not be good way therefore.But when each object to be identified has only a width of cloth training image, be difficult to originally be utilized, and instead made the notable attribute of perspective in follow-up principal component analysis (PCA) process, can more show especially out and abandoned them by these information of being abandoned; On the other hand, this Fuzzy processing makes that also to add projected image insensitive to the little variation of expression, illumination etc., thereby makes the inventive method can carry out the recognition of face of multiple expression, many illumination conditions to a certain extent.Therefore, the perspective view among Fig. 3 and add the core that projected image is the inventive method.

Claims (2)

1, a kind of automatic human face identification method based on everyone piece image, comprise that digital facial image detects and recognition device obtains gray level image by Digital Image Input Device, utilize the information of eye location to obtain human face region, then facial image is carried out feature extraction and compares, it is characterized in that this method may further comprise the steps with the existing image in the facial image database:

(1) if recognition mechanism does not train, execution in step (2) then, otherwise forward step (4) to;

(2) generate the projected image that adds of training image;

(3) utilize the principal component analysis (PCA) technology to generate the feature space that adds projected image;

(4) receive image to be identified;

(5) generate the projected image that adds of image to be identified;

(6) projected image that adds that adds projected image and training image for the treatment of recognition image in adding the projected image feature space is compared;

(7) with the pairing object of training image of comparison similarity maximum as recognition result;

(8) finish.

2, automatic human face identification method according to claim 1 is characterized in that:

In (2), (5), the step that generation adds projected image is:

(1) obtains gray level image;

(2) calculate the vertical projection V of image by following formula _P(x) and horizontal projection H _P(y), wherein P is a width of cloth N ₁* N ₂The gray level image of size, P (x, y) expression P mid point (x, the gray-scale value of y) locating, x ∈ [1, N ₁], y ∈ [1, N ₂]: $V_P\left(x\right)=\underset{y=1}{\overset{N_2}{\mathrm{\Σ}}}P(x,y)$ $H_P=\left(y\right)=\underset{x=1}{\overset{N_1}{\mathrm{\Σ}}}P(x,y);$

(3) the perspective view M of generation original image (being P) _P, M _PAt point (x, the gray-scale value M that y) locates _P(x, y) calculate by following formula: $M_P=(x,y)=\frac{V_P\left(x\right)H_P\left(y\right)}{N_1{N}_2\stackrel{\‾}{P}},$ Wherein P is the average gray value of P: $\stackrel{\‾}{P}=\frac{\underset{x=1}{\overset{N_1}{\mathrm{\Σ}}}\underset{y=1}{\overset{N_2}{\mathrm{\Σ}}}P(x,y)}{N_1{N}_2};$

What (4) generate original image adds projected image P _α, P _αAt point (x, the gray-scale value P that y) locates _α(x, y) calculate by following formula: $P_{\mathrm{\α}}(x,y)=\frac{P(x,y)+\mathrm{\α}{M}_P(x,y)}{1+\mathrm{\α}}$ Wherein α is the fractional value for example 0.25 between (0,1) of user preset.

Images