CN102360421B - Face identification method and system based on video streaming - Google Patents

Abstract

The invention discloses a face identification method and system based on video streaming. The method comprises the steps of: receiving a video streaming to be identified acquired by a video acquisition device; carrying out face detection on each frame of image in the video streaming to be identified to determine a face image to be identified; positioning a characteristic point corresponding of each frame of face image to be identified; determining a key face image in the face image to be identified; determining a key characteristic point of corresponding characteristic points in each frame of key face image; carrying out image pre-processing on the determined key face image; and determining a face identification result corresponding to the video streaming to be identified according to a weighting processing result of a similarity of the key characteristic point in each frame of key face image and a characteristic point corresponding to each face image in a face image database. In the scheme, through manners of detecting a color histogram-based key frame, determining the key characteristic point and weighting the similarity, identification offsets brought by the influence of a video acquisition environment are eliminated, and the face identification is carried out rapidly and effectively.

Description

A kind of face identification method and system based on video flowing

Technical field

The present invention relates to simulation technical field, particularly relate to a kind of face identification method and system based on video flowing.

Background technology

Recognition of face is the biological identification technology that a kind of face feature information based on people is carried out identification, its with respect to have as other more ripe method for identifying human body biological characteristics such as fingerprint, DNA detections non-direct infringement, overcast, interaction strong, the advantage such as be convenient to follow the tracks of afterwards, therefore, be for many years a study hotspot always.Recognition of face is image or the video flowing that contains face with the collection of the equipment such as camera or video camera, automatically detection and tracking face in image, and then the face detecting is carried out to a series of correlation techniques of face, comprise man face image acquiring, face location, recognition of face pre-service, memory storage and contrast identification, reach the object of identification different people identity.

Wherein, the recognition of face based on video flowing refers to and is input as face video, and utilizes rest image face database identify or verify.Owing to conventionally comprising more information in video flowing, for example: the facial image in multiple image, the video of same person has continuity on time and space, estimated three-dimensional face structure, recovered high-definition picture etc. from low-resolution image by motion change, can prevent identification deception based on still image etc.

Therefore, the recognition of face based on video flowing can utilize more information in video, more has superiority than traditional recognition of face based on rest image, becomes the area research personnel's such as criminal's identification, immigration management, housekeeping robot concern.

Summary of the invention

For solving the problems of the technologies described above, the embodiment of the present invention provides a kind of face identification method and system based on video flowing, and technical scheme is as follows:

Based on a face identification method for video flowing, comprising:

The video flowing to be identified that receiver, video collecting device gathers;

Every two field picture in described video flowing to be identified is carried out to face detection, to determine facial image to be identified;

Locate every frame facial image characteristic of correspondence point to be identified;

According to the corresponding color histogram of facial image to be identified, determine the crucial facial image in facial image to be identified;

Determine the key feature points in the crucial facial image character pair point of every frame;

Determined crucial facial image is carried out to image pre-service, to reduce image geometry feature and the illumination impact on crucial facial image;

According to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determine face recognition result corresponding to described video flowing to be identified.

Accordingly, the present invention also provides a kind of face identification system based on video flowing, comprising:

Video reception module, the video flowing to be identified gathering for receiver, video collecting device;

Face detection module, for the every two field picture of described video flowing to be identified is carried out to face detection, to determine facial image to be identified;

Positioning feature point module, for locating every frame facial image characteristic of correspondence point to be identified;

Key frame determination module, for according to the corresponding color histogram of facial image to be identified, determines the crucial facial image in facial image to be identified;

Key point determination module, for determining the key feature points of the crucial facial image character pair point of every frame;

Pretreatment module, for determined crucial facial image is carried out to image pre-service, to reduce image geometry feature and the illumination impact on crucial facial image;

Result determination module, according to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determines face recognition result corresponding to described video flowing to be identified.

In the technical scheme that the embodiment of the present invention provides, the video flowing that video capture device is gathered carries out after face detection and positioning feature point, utilize the color histogram of facial image, determine the crucial facial image in facial image, and then crucial facial image is reduced to the image pre-service of geometric properties and illumination effect and determining of key feature points, last according to the weighted value of the key feature points similarity corresponding with each faceform in face database, determine final face recognition result.In this programme, the key frame by based on color histogram detects and key feature points is determined, the mode of Similarity-Weighted, eliminates the identification deviation of bringing due to video acquisition environmental impact, fast and effeciently carries out recognition of face.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

The first process flow diagram of a kind of face identification method based on video flowing that Fig. 1 provides for the embodiment of the present invention;

The second process flow diagram of a kind of face identification method based on video flowing that Fig. 2 provides for the embodiment of the present invention;

Fig. 3 is the schematic diagram to face vertical integral projection in the embodiment of the present invention;

Fig. 4 is the schematic diagram to the horizontal integral projection of face in the embodiment of the present invention;

Fig. 5 is the second schematic diagram to face vertical integral projection in the embodiment of the present invention;

Fig. 6 is image rotation schematic diagram in the embodiment of the present invention;

Fig. 7 is the histogram schematic diagram before crucial face image equalization in the embodiment of the present invention;

Fig. 8 is to the histogram schematic diagram after crucial face image equalization in the embodiment of the present invention;

A kind of the third process flow diagram of the face identification method based on video flowing that Fig. 9 provides for the embodiment of the present invention;

The structural representation of a kind of face identification system based on video flowing that Figure 10 provides for the embodiment of the present invention.

Embodiment

The embodiment of the present invention provides a kind of face identification method and system based on video flowing, carries out recognition of face fast and effectively with the video flowing of realizing under the general application scenarios that utilization gathered.First a kind of face identification method based on the video flowing embodiment of the present invention being provided is below introduced.

Based on a face identification method for video flowing, comprising:

The video flowing to be identified that receiver, video collecting device gathers;

Every two field picture in described video flowing to be identified is carried out to face detection, to determine facial image to be identified;

Locate every frame facial image characteristic of correspondence point to be identified;

According to the corresponding color histogram of facial image to be identified, determine the crucial facial image in facial image to be identified;

Determine the key feature points in the crucial facial image character pair point of every frame;

Determined crucial facial image is carried out to image pre-service, to reduce image geometry feature and the illumination impact on crucial facial image;

According to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determine face recognition result corresponding to described video flowing to be identified.

In the technical scheme that the embodiment of the present invention provides, the video flowing that video capture device is gathered carries out after face detection and positioning feature point, utilize the color histogram of facial image, determine the crucial facial image in facial image, and then crucial facial image is reduced to the image pre-service of geometric properties and illumination effect and determining of key feature points, last according to the weighted value of the key feature points similarity corresponding with each faceform in face database, determine final face recognition result.In this programme, the key frame by based on color histogram detects and key feature points is determined, the mode of Similarity-Weighted, eliminates the identification deviation of bringing due to video acquisition environmental impact, fast and effeciently carries out recognition of face.

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, rather than whole embodiment.Based on the embodiment in the present invention, those of ordinary skills, not making the every other embodiment obtaining under creative work prerequisite, belong to the scope of protection of the invention.

As shown in Figure 1, a kind of face identification method based on video flowing, can comprise:

S101, the video flowing to be identified that receiver, video collecting device gathers;

Conventionally video capture device generally can comprise: video camera, video recorder, LD video machine etc.These video capture devices are arranged on needs the region of video acquisition ad-hoc location.In the time that the video flowing that need to gather video capture device carries out analyzing and processing, it is connected with corresponding analyzing and processing equipment.

S102, carries out face detection to every two field picture in described video flowing to be identified, to determine facial image to be identified;

In each two field picture region of received video flowing to be identified, may not exist face, human face ratio less, or have multiple uncorrelated faces, complete face etc.That is to say, the information that each two field picture in video flowing to be identified comprises not is all available, so need to carry out face detection to the every two field picture in video flowing to be identified, by comprise complete face and meet certain human face ratio picture frame extract, carry out follow-up recognition of face processing.

Wherein, as shown in Figure 2, every two field picture in described video flowing to be identified is carried out to face detection, is specifically as follows:

S201, carries out gray processing and histogram equalization processing to every two field picture in described video flowing to be identified;

S202, the face that every two field picture after treatment is carried out to multiple dimensioned many features detects;

S203, carries out many features by the corresponding testing result of each yardstick of every two field picture and merges processing;

S204, merges result by many features corresponding every two field picture and carries out multiple dimensioned merging processing;

S205, is defined as facial image to be identified by the multiple dimensioned merging result that meets default face yardstick threshold value.

In above-mentioned processing procedure, each two field picture carries out after gray processing and histogram equalization process in to video flowing to be identified, and the face that every two field picture is carried out respectively to multiple dimensioned many features detects, to determine the testing result for different scale different characteristic.So-called multiple dimensioned be characterized as: every two field picture is carried out to the detection of each yardstick, and in the time that each yardstick is detected, detect according to different features more.Carrying out after multiple dimensioned many feature detection, testing result is being carried out to the multiple dimensioned merging processing of many features, to eliminate the deviation of different testing results, the corresponding amalgamation result of final every two field picture.Be understandable that, if contained face information is less in amalgamation result, this two field picture be not suitable for recognition of face.So, the multiple dimensioned merging result that meets default face yardstick threshold value need to be defined as to facial image to be identified.Be understandable that, face yardstick threshold value can be set according to actual conditions, does not limit at this.

S103, locates every frame facial image characteristic of correspondence point to be identified;

Be understandable that, face is made up of parts such as eyes, nose, face, chins, and the shape of these parts, size and structural each species diversity just make each face vary, therefore, for the accurate location of these organs, be a vital step for whole face recognition process.Wherein, because eyes are sockdolager's face features relatively in the middle of face, so as long as eyes are accurately positioned, other organs of face, as: eyebrow, face, nose etc., can be located more accurately by potential distribution relation.

In the present embodiment, the location of features of human face images to be identified can be undertaken by the crest corresponding to producing under different integral projection modes or trough.Wherein, integral projection is divided into vertical projection and horizontal projection, establishes the gray-scale value that f (x, y) presentation video (x, y) is located, at image [y ₁, y ₂] and [x ₁, x ₂] the horizontal integral projection M in region _hand vertical integral projection M (y) _v(x) be expressed as:

$M_h\left(y\right)=\frac{1}{x_2-x_1}\underset{x_1}{\overset{x_2}{\mathrm{\Σ}}}G(x,y)---\left(1\right)$

$M_v\left(x\right)=\frac{1}{y_2-y_1}\underset{y_1}{\overset{y_2}{\mathrm{\Σ}}}G(x,y)---\left(2\right)$

Wherein, horizontal integral projection is exactly to show after the gray-scale value of all pixels of a line is added up again; And vertical integral projection is exactly to show after the gray-scale value of all pixels of row is added up again.

Be understandable that, before carrying out positioning feature point, in order to obtain the obvious drop shadow curve of grey scale change, can increase the contrast of image, for example, adopt adaptive threshold method to carry out after binary conversion treatment, then carry out integral projection.

For example: as shown in Fig. 3 (a), Fig. 3 (b) and Fig. 3 (c), for to the vertical integral result of facial image, from each drop shadow curve, can find out, at face left and right boundary, the gray scale of image has greatly changed, and border, left and right correspondence two larger troughs on both sides in curve.Therefore, only need these two trough point x of location ₁, x ₂from this facial image to be identified transverse axis [x ₁, x ₂] region image interception out, can realize the location on border, facial image to be identified left and right.To binaryzation facial image to be identified after the boundary alignment of left and right, carry out respectively horizontal integral projection and vertical integral projection, its projection result is respectively as shown in Figure 4, Figure 5.Further, utilize knownly to the priori of facial image, eyebrow and eyes are black regions nearer in facial image, and its correspondence the first two minimum point in horizontal integral projection curve.As shown in Figure 4, what first minimum point was corresponding is the position of eyebrow on the longitudinal axis, and note is y _brow, what second minimum point was corresponding is the positions of eyes on the longitudinal axis, note is y _eye, what the 3rd minimum point was corresponding is the position of nose on the longitudinal axis, note is y _nose, what the 4th minimum point was corresponding is the position of face on the longitudinal axis, note is y _month.Equally, as shown in Figure 5, there are two minimum points in facial image central symmetry axis both sides, the position of corresponding right and left eyes on transverse axis respectively, and note is x _left-eye, x _right-eye; The position of eyebrow on transverse axis is identical with eyes; Face and the nose position on transverse axis is (x _left-eye+ x _right-eye)/2.

Be understandable that, the locator meams of the unique point of facial image to be identified is not limited to the mode described in the present embodiment.

S104, according to the corresponding color histogram of facial image to be identified, determines the crucial facial image in facial image to be identified;

Due in video to be identified, same person can occur in a lot of frames, in a frame, also having a lot of people occurs, owing to affected by attitude, illumination variation, front, side, dimensional variation, expression shape change etc., its picture quality is different, in order to improve accuracy rate and the efficiency of identification, need to filter out the good image of some mass ratioes, carry out follow-up recognition of face, good filtered out quality each two field picture is defined as to crucial facial image.Namely, under normal circumstances, many for the video picture frame of recognition of face, if each facial image is all processed, must affect running efficiency of system, reduce system real time.Meanwhile, include bulk redundancy frame in face video, comprise and rotate the excessive and even facial image frame of uneven illumination, these frames are carried out to feature location can be very difficult, even locates unsuccessfully, directly affects the discrimination of entire system.

In this example, according to the corresponding color histogram of facial image to be identified, determine the key frame images in facial image to be identified.

Wherein, color histogram has reflected the probability that in piece image, pixel color occurs, is the estimation of pixel probability of occurrence.A given width digital picture I, its color histogram vector H can be expressed as:

$H=\left(h\right[c_1],h[c_2],\·\·\·,h[c_k],\·\·\·,h[c_N\left]\right),\underset{k=1}{\overset{N}{\mathrm{\Σ}}}h\left[c_k\right]=\mathrm{1,0}\≤h\left[c_k\right]\≤1---\left(3\right)$

Wherein h[c _k] be the probability number that k kind color occurs in this digital picture:

$h\left[c_k\right]=\frac{\underset{i=0}{\overset{N_1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N_2}{\mathrm{\&Sigma;}}}\left\{\begin{array}{c}1,I(i,j)={\mathrm{<figure-callout\; id="0"\; label="c\; k"\; filenames="HDA0000100220370000041.png,HDA0000100220370000051.png"\; state="\{\{state\}\}">c</figure-callout>}}_k\\ 0,\mathrm{otherwise}\end{array}\right.}{N_1*N_2}---\left(4\right)$

N ₁, N ₂presentation video has N respectively ₁row N ₂row, the pixel value that I (i, j) presentation video mid point (i, j) is located.

Color histogram has reflected the global information of image in color, and every width image has the color histogram of answering in contrast.If G and H are the color histogram vector of the two width images that will compare, the color progression occurring in N presentation video, g _k, h _kthe frequency that in the color histogram of presentation video G and H, k level look occurs respectively, the similarity of two width images can represent with the Euclidean distance between its color histogram:

$d(G,H)=\frac{1}{H}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{(g_k-h_k)}^{\frac{1}{2}}---\left(5\right)$

Color histogram, as a kind of very important visual feature, can effectively be distinguished the object of different structure, size, shape, and color identification is simultaneously a kind of both simple and quick method, and this is to there being the system of requirement of real-time particularly important.Based on this, the present invention proposes a kind of face key frame of video extraction algorithm based on color histogram, and before feature extraction and identification, advanced pedestrian's face key frame detects, and avoids redundant frame to process, thereby greatly improves system performance.Detailed process is as follows:

In face video sequence, select the good two field picture of quality as standard faces image, calculate its color histogram H, as standard faces color histogram;

Obtain frame by frame the facial image to be identified in face video sequence, calculate its color histogram G _i, and and the comparison of Standard Colors histogram, through type (5) calculates its Euclidean distance;

Setpoint distance threshold value T (0≤T≤1), the facial image all to be identified that Euclidean distance is less than to T is as crucial facial image, and the facial image to be identified that Euclidean distance is greater than distance threshold is defined as to non-key facial image.Wherein, T can control the quality and quantity of choosing crucial facial image.

Be understandable that, carry out follow-up recognition of face by filter out crucial facial image in facial image to be identified, can effectively improve system treatment effeciency, more there is practicality simultaneously.

S105, determines the key feature points in the crucial facial image character pair point of every frame;

Be understandable that, face different parts unique point is concerning recognition of face, differentiated, namely, not all unique point all has higher identification value,, for example, find according to experimental result, the recognition capability of the genius loci points such as eyebrow, face is higher than positions such as eyes, nose, profiles.Therefore, need to determine the key feature points in the crucial facial image character pair point of every frame, to carry out effective recognition of face.

S106, carries out image pre-service to determined crucial facial image, to reduce image geometry feature and the illumination impact on crucial facial image;

Wherein, determined key frame is carried out to geometrical normalization pre-service, namely, to facial image process translation, rotation and filtering processing, the identification deviation of being brought to reduce image geometry feature, is convenient to feature extraction and identification.To the normalized processing procedure of determined crucial face image geometry can be:

The position coordinates of supposing two eyes in crucial facial image is respectively E ₁(x ₁, y ₁) and E _r(x _r, y _r), the distance of eye center is d, and the angle of two axis oculi lines and x axle is θ, and θ is the angle value that will rotate.

$d=\sqrt{{(x_r-x_l)}^2+{(y_r-y_l)}^2}---\left(6\right)$

$\mathrm{\&theta;}=\arctan \left(\frac{y_r-y_l}{x_r-x_l}\right)---\left(7\right)$

According to the priori of facial image ratio, the center C (x of face _c, y _c) be:

$x_c=\frac{x_r-x_l}{2}+d\&times;\sin \left(\mathrm{\&theta;}\right)$ (8)

$y_c=\frac{y_r-y_l}{2}+d\&times;\cos \left(\mathrm{\&theta;}\right)$

, centered by C, take θ as the anglec of rotation, face is rotated, the computing formula of postrotational face coordinate (x ', y ') is:

x′＝x _c+cosθ(x-x _c)+sinθ(y-y _c)

(9)

y′＝y _c+cosθ(y-y _c)+sinθ(x-x _c)

Key frame images rotation schematic diagram as shown in Figure 6.

Be understandable that, the image changing through over-rotation may size differ, so also need regular facial image to unified size.Wherein the size normalization of image adopts the method for convergent-divergent to realize conventionally, and concrete processing procedure can be:

The conversion of pixel coordinate, is mapped to output image by the pixel of input picture.If the axial scaling of image x is R _x, the axial scaling of y is R _y, the transformation matrix of image scaling is as follows:

$\left[\begin{array}{c}{x}^{\&prime;}\\ y^{\&prime;}\\ 1\end{array}\right]=\left[\begin{array}{ccc}{R}_x& 0& 0\\ 0& {\mathrm{<figure-callout\; id="0"\; label="R\; y"\; filenames="HDA0000100220370000041.png,HDA0000100220370000051.png"\; state="\{\{state\}\}">R</figure-callout>}}_y& 0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{c}x\\ y\\ 1\end{array}\right]---\left(10\right)$

Wherein, in the time that image is amplified, the pixel in output image may can not find corresponding pixel in source images, therefore, must carry out interpolation arithmetic to it.Conventional interpolation arithmetic method comprises: arest neighbors interpolation, bilinear interpolation and bicubic interpolation.Arest neighbors interpolation is to obtain with the picture element interpolation of the several points in its neighborhood, and algorithm is simple, and operand is very little, but after conversion, gray scale has obvious uncontinuity; Bicubic interpolation computational accuracy is high, but calculated amount is larger.Therefore, conventional bilinear interpolation compensates.Postulated point (x ₁, y ₁), (x ₁, y ₂), (x ₂, y ₁), (x ₂, x ₂) be four summits of rectangular area in image, point (x, y) is included in this rectangle, and the computing method of the gray-scale value of (x, y) are as follows:

f(x，y ₀)＝f(x ₀，y ₀)+(x-x ₀)/(x ₁-x ₀)[f(x ₁，y ₀)-f(x ₀，y ₀)]

f(x，y ₁)＝f(x ₀，y ₁)+(x-x ₀)/(x ₁-x ₀)[f(x ₁，y ₁)-f(x ₀，y ₁)] (11)

f(x，y)＝f(x，y ₀)+(y-y ₀)/(y ₁-y ₀)[f(x，y ₁)-f(x，y ₀)]

Further, be one of principal element affecting face recognition result due to illumination, therefore need to eliminate the impact of illumination on facial image.In the present embodiment, can adopt histogram equalization techniques to carry out image irradiation pre-service, principle is the gray level by changing each pixel in image, changes the histogram of image with this, thereby makes partially bright image dimmed or partially dark image is brightened.Grey level histogram is the statistics of each grey scale pixel value occurrence number or frequency in piece image, is the function of gray-scale value, is a instrument simple, the most useful in Digital Image Processing.It has not only described the gray shade scale content of piece image, and has described the number in image with the pixel of this gray-scale value.The histogram of any piece image has all comprised considerable information, and the image of some type can also be described completely by its histogram.

Wherein, concrete disposal route can comprise:

Suppose that number of pixels total in gray level image is n, a certain gray level is r _knumber of pixels be n _k, n _k/ n be exactly gray level be r _kthe probability that occurs of pixel, add up the probability that in this gray level image, all gray levels occur, can obtain the grey level histogram of this image.If with pixel grayscale in variable r representative image, and r is done to normalized: 0≤r≤1, wherein, r=0 represents black, r=1 representative is white.For a given image, the gray level that each pixel obtains in [0,1] interval is random.R is a stochastic variable, can use probability density function P _r(r) represent the intensity profile of original image, can find out the gray distribution features of piece image from the distribution of image gray levels.If the gray-scale value of most of pixels is taken at the region close to 0 in image, entire image is darker, otherwise, entire image trial of strength.The gray-scale value right and wrong of many images are equally distributed, and it is very common that gray-scale value concentrates on an image in minizone.

The object of histogram equalization processing is that to make the histogram of image after treatment be straight, and each gray level has the identical frequency of occurrences, and each gray level has uniform probability distribution thus, thereby improves the subjective quality of image.Therefore, need to find a kind of transformation relation S=T (r), make the rear new grey level histogram of conversion convert front histogram straight.Can adopt the cumulative distribution function of r as transfer function, that is:

$S=T\left(r\right)={\&Integral;}_0^r{P}_r\left(w\right)\mathrm{dw}---\left(12\right)$

Wherein, w is integration variable, it is the cumulative distribution function of r.Use r _krepresent discrete gray-scale value, use P _r(r _k) represent P _r(r), have:

$P_r\left(r_k\right)=\frac{{\mathrm{<figure-callout\; id="0"\; label="n\; k\; n"\; filenames="HDA0000100220370000041.png,HDA0000100220370000051.png"\; state="\{\{state\}\}">n</figure-callout>}}_k}{n}$ 0≤r _k≤1，k＝0，1，2…，L-1 (13)

Wherein, n _kfor occurring the number of picture elements of this gray scale in image, n is pixel sum in image, and L presentation video gray-scale value is divided into L level.

The discrete form of transfer function is:

$S_k=T\left(r_k\right)=\underset{j=0}{\overset{k}{\mathrm{\&Sigma;}}}\frac{n_j}{n}=\underset{j=0}{\overset{k}{\mathrm{\&Sigma;}}}{P}_r\left(r_j\right)$ 0≤r _k≤1，k＝0，1，2…，L-1 (14)

Take any width facial image as example, Fig. 7 is the schematic diagram before histogram equalization, and Fig. 8 is the schematic diagram after histogram equalization.Visible, the intensity of illumination of the image after histogram equalization has obtained compensation, and gray-scale value dynamic range increases, and intensity profile is more even, and integral image contrast strengthens.

S107, according to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determines face recognition result corresponding to described video flowing to be identified.

Be understandable that, in the time carrying out recognition of face, first should form corresponding face database by features training mode, think that recognition of face provides discrimination standard.In the process of carrying out features training, represent face using deformable two grids as template.Extract two grids at face area, these grids can be regarded as a two-dimensional topology figure, for each node in topological diagram, calculate its characteristic information, and give this node by obtained value and obtain labeled graph, be specially and first choose the special point (such as eyebrow, eyes, nose, chin etc.) in position on facial image as unique point; Then by Gabor wave filter, unique point is carried out to filtering, extract Gabor coefficient composition eigenvector, represent unique point and the characteristic of correspondence vector thereof of each face and be stored in database with face figure.

For example: the data in 1000 people's training storehouse are exactly the raw data of 1000 people's human face characteristic point.On facial image, the special point (such as eyebrow, eyes, nose, chin etc.) in position is as unique point; Then by Gabor wave filter, unique point is carried out to filtering, extract Gabor coefficient and form eigenvector and form face figure, represent unique point and the characteristic of correspondence vector thereof of each face and be stored in to train in storehouse with face figure.

Wherein, described according to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determine face recognition result corresponding to described video flowing to be identified, as shown in Figure 9, be specifically as follows:

S901, utilizes wavelet transformation to determine the corresponding critical eigenvalue of each key feature points;

The feature extraction of face information is a vital step in recognition of face, whether the characteristic information extracting is stable, reliable, abundant etc. all by the discrimination final system that has influence on, particularly in the time there is the variations such as illumination, expression, attitude and other interference in facial image.Two wavelet transformations have locality, directional selectivity and frequency selectivity, can accurately extract the local feature of different directions in facial image, frequency and yardstick, and have certain antijamming capability, are therefore widely used in the feature extraction of face.Wherein, the functional form of 2-d wavelet can be expressed as:

${\mathrm{\&psi;}}_j\left(\stackrel{\&RightArrow;}{x}\right)=\frac{{\left|\right|\stackrel{\&RightArrow;}{k_j}\left|\right|}^2}{{\mathrm{\&sigma;}}^2}\exp (-\frac{{\left|\right|\stackrel{\&RightArrow;}{k_j}\left|\right|}^2{\left|\right|\stackrel{\&RightArrow;}{x}\left|\right|}^2}{2{\mathrm{\&sigma;}}^2})[\exp \left(i\stackrel{\&RightArrow;}{k_j}\stackrel{\&RightArrow;}{x}\right)-\exp (-\frac{{\mathrm{\&sigma;}}^2}{2})]---\left(15\right)$

Wherein wave vector

for

In formula,

for the image coordinate of given position; also claim the centre frequency of wave filter;

reflect the directional selectivity of wave filter.In natural image,

being used for compensating the energy spectrum being determined by frequency decays;

it is the Gaussian envelope function for constraint plane ripple;

for complex values plane wave, its real part is cosine plane wave

imaginary part is sinusoidal plane wave

be called flip-flop compensation, be used for the impact on Two-Dimensional Gabor Wavelets of the flip-flop of removal of images, make Two-Dimensional Gabor Wavelets conversion not be subject to the impact of gradation of image absolute figure, and insensitive to the illumination variation of image.The function of two-dimensional Gabor filter is a complex function, and its real part and imaginary part can be expressed as respectively:

$R\left({\mathrm{\&psi;}}_j\left(\stackrel{\&RightArrow;}{x}\right)\right)=\frac{{\left|\right|\stackrel{\&RightArrow;}{k_j}\left|\right|}^2}{{\mathrm{\&sigma;}}^2}\exp (-\frac{{\left|\right|\stackrel{\&RightArrow;}{k_j}\left|\right|}^2{\left|\right|\stackrel{\&RightArrow;}{x}\left|\right|}^2}{2{\mathrm{\&sigma;}}^2})[\cos \left(\stackrel{\&RightArrow;}{k_j}\stackrel{\&RightArrow;}{x}\right)-\exp (-\frac{{\mathrm{\&sigma;}}^2}{2})]---\left(17\right)$

$I\left({\mathrm{\&psi;}}_j\left(\stackrel{\&RightArrow;}{x}\right)\right)=\frac{{\left|\right|\stackrel{\&RightArrow;}{k_j}\left|\right|}^2}{{\mathrm{\&sigma;}}^2}\exp (-\frac{{\left|\right|\stackrel{\&RightArrow;}{k_j}\left|\right|}^2{\left|\right|\stackrel{\&RightArrow;}{x}\left|\right|}^2}{2{\mathrm{\&sigma;}}^2})\left[\sin \left(\stackrel{\&RightArrow;}{k_j}\stackrel{\&RightArrow;}{x}\right)\right]---\left(18\right)$

Two-Dimensional Gabor Wavelets conversion has been described in image I more given the gray feature of near zone, can realize by the convolution of Gabor family of functions and image the filtering of image I:

$J_{j,{\stackrel{\&RightArrow;}{x}}_0}\left({\stackrel{\&RightArrow;}{x}}_0\right)=\&Integral;I\left(\stackrel{\&RightArrow;}{x}\right){\mathrm{\&psi;}}_j({\stackrel{\&RightArrow;}{x}}_0-\stackrel{\&RightArrow;}{x})d^{2}x---\left(19\right)$

Two-dimensional Gabor filter is bandpass filter, all have good resolution characteristic in spatial domain and frequency field, and its parameter has embodied its sample mode in spatial domain and frequency field, has determined its ability to express to signal.Conventionally can adopt the Gabor bank of filters with multiple centre frequencies and direction to carry out Description Image.Parameter kv,

different choice embodied respectively the sample mode of Two-Dimensional Gabor Wavelets at frequency and director space; Parameter σ determines the bandwidth of wave filter.The experiment of Lades shows: for the image that is of a size of 128 × 128, as wave filter center of maximum frequency k _maxfor pi/2, when the π of σ=2, reach best experimental result.Because the texture of image is stochastic distribution,

span be 0～2 π, if consider the symmetry of Gabor wave filter,

actual span be 0 to π.For the local feature of Description Image, the present embodiment adopts 40 Gabor wave filters of 5 centre frequencies, 8 direction compositions to carry out filtering to image.Parameter k _vwith

value as follows:

In formula, f is the interval factor for limiting frequency domain Kernel Function distance, is conventionally taken as

v ∈ { 0,1,2,3,4}, u={0,1,2,3,4,5,6,7}, j=u+8v.

S902, calculates in the crucial facial image of every frame the similarity of each faceform's character pair point in each key feature points and face database;

S903, is weighted the similarity of each key feature points and same faceform's character pair point in crucial every frame facial image, faceform corresponding maximum weighted value is defined as to the alternate recognition results of described crucial facial image;

S904, adds up crucial face amount of images corresponding to each alternative face recognition result, and by quantity at most and be defined as higher than the alternative face recognition result of predetermined number threshold value the face recognition result that video flowing to be identified is corresponding.

In above-mentioned processing, by calculating the similarity of each key feature points and the character pair point in each faceform of presetting in the crucial facial image of every frame, all key feature points in the crucial facial image of each frame are weighted for same faceform's similarity, and faceform corresponding maximum weighted value are defined as to the alternate recognition results of described crucial facial image; Then, add up crucial face amount of images corresponding to each alternative face recognition result, and by quantity at most and be defined as higher than the alternative face recognition result of predetermined number threshold value the face recognition result that video flowing to be identified is corresponding.Be understandable that, in actual applications, in the video flowing gathering, might not comprise certain faceform in face database, therefore, crucial face amount of images corresponding to a certain alternative face recognition result may be at most, but when close with other alternative face recognition result quantity, show that gathered video flowing may not include the faceform in face database.Therefore, need to set an amount threshold, to improve the accuracy of identification.

By the description of above embodiment of the method, those skilled in the art can be well understood to the mode that the present invention can add essential general hardware platform by software and realize, can certainly pass through hardware, but in a lot of situation, the former is better embodiment.Based on such understanding, the part that technical scheme of the present invention contributes to prior art in essence in other words can embody with the form of software product, this computer software product is stored in a storage medium, comprise that some instructions (can be personal computers in order to make a computer equipment, server, or the network equipment etc.) carry out all or part of step of method described in each embodiment of the present invention.And aforesaid storage medium comprises: various media that can be program code stored such as ROM (read-only memory) (ROM), random access memory (RAM), magnetic disc or CDs.

Corresponding to embodiment of the method above, the embodiment of the present invention also provides a kind of face identification system based on video flowing, as shown in figure 10, can comprise:

Video reception module 110, the video flowing to be identified gathering for receiver, video collecting device;

Face detection module 120, for the every two field picture of described video flowing to be identified is carried out to face detection, to determine facial image to be identified;

Positioning feature point module 130, for locating every frame facial image characteristic of correspondence point to be identified;

Key frame determination module 140, for according to the corresponding color histogram of facial image to be identified, determines the crucial facial image in facial image to be identified;

Key point determination module 150, for determining the key feature points of the crucial facial image character pair point of every frame;

Pretreatment module 160, for determined crucial facial image is carried out to image pre-service, to reduce image geometry feature and the illumination impact on crucial facial image;

Result determination module 170, according to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determines face recognition result corresponding to described video flowing to be identified.

Wherein, described face detection module comprises:

Equilibrium treatment unit, carries out gray processing and histogram equalization processing to every two field picture in described video flowing to be identified;

Face detecting unit, detects for the face that every two field picture after treatment is carried out to multiple dimensioned many features;

Many features merge cells, merges processing for the corresponding testing result of each yardstick of every two field picture being carried out to many features;

Multiple dimensioned merge cells, carries out multiple dimensioned merging processing for many features corresponding every two field picture are merged to result;

Result determining unit, for being defined as facial image to be identified by the multiple dimensioned merging result that meets default face yardstick threshold value.

Wherein, described positioning feature point module comprises:

Enhancement contrast unit, for adopting adaptive threshold method to carry out Enhancement contrast processing to facial image to be identified;

Positioning feature point unit, utilizes vertical integral projection method and horizontal integral projection method to carry out positioning feature point processing to facial image to be identified, to locate described facial image characteristic of correspondence point to be identified.

Wherein, described key frame determination module comprises:

Standard unit, for the facial image to be identified that settles the standard, and calculates the color histogram of the corresponding human face region of described standard facial image to be identified;

Color histogram determining unit, for calculating the color histogram of the corresponding human face region of every frame facial image to be identified;

Histogram difference computing unit, for calculating the Euclidean distance of the color histogram that color histogram that current facial image to be identified is corresponding is corresponding with standard facial image to be identified;

Key frame determining unit, for judging whether described Euclidean distance is less than predeterminable range threshold value, if so, is defined as crucial facial image by described current facial image to be identified; If not, described facial image to be identified is defined as to non-key facial image.

Wherein, described pretreatment module comprises:

Geometrical normalization unit, for carrying out translation, rotation and/or filtering processing to described crucial facial image, to realize geometrical normalization processing;

Illumination pretreatment unit, for the crucial facial image through above-mentioned processing is carried out to histogram equalization processing, to realize illumination pretreatment.

Wherein, described result determination module comprises:

Eigenwert computing unit, for utilizing wavelet transformation to determine the corresponding critical eigenvalue of each key feature points;

Similarity calculated, calculates in the crucial facial image of every frame the similarity of each faceform's character pair point in each key feature points and face database;

Alternative result determining unit, is weighted the similarity of each key feature points and same faceform's character pair point in crucial every frame facial image, faceform corresponding maximum weighted value is defined as to the alternate recognition results of described crucial facial image;

Recognition result determining unit, for adding up crucial face amount of images corresponding to each alternative face recognition result, and by quantity at most and be defined as higher than the alternative face recognition result of predetermined threshold value the face recognition result that video flowing to be identified is corresponding.

For device or system embodiment, because it is substantially corresponding to embodiment of the method, so relevant part is referring to the part explanation of embodiment of the method.Device described above or system embodiment are only schematic, the wherein said unit as separating component explanation can or can not be also physically to separate, the parts that show as unit can be or can not be also physical locations, can be positioned at a place, or also can be distributed in multiple network element.Can select according to the actual needs some or all of module wherein to realize the object of the present embodiment scheme.Those of ordinary skills, in the situation that not paying creative work, are appreciated that and implement.

In several embodiment provided by the present invention, should be understood that, disclosed system, apparatus and method, not exceeding in the application's spirit and scope, can realize in other way.Current embodiment is a kind of exemplary example, should not serve as restriction, and given particular content should in no way limit the application's object.For example, the division of described unit or subelement, is only that a kind of logic function is divided, and when actual realization, can have other dividing mode, and for example multiple unit or multiple subelement combine.In addition, multiple unit can or assembly can in conjunction with or can be integrated into another system, or some features can ignore, or do not carry out.

In addition, institute's descriptive system, the schematic diagram of apparatus and method and different embodiment, not exceeding in the application's scope, can with other system, module, technology or method in conjunction with or integrated.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be by some interfaces, indirect coupling or the communication connection of device or unit can be electrically, machinery or other form.

The above is only the specific embodiment of the present invention; it should be pointed out that 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 (8)

1. the face identification method based on video flowing, is characterized in that, comprising:

The video flowing to be identified that receiver, video collecting device gathers;

Every two field picture in described video flowing to be identified is carried out to face detection, to determine facial image to be identified;

By crest or trough corresponding to producing under different integral projection modes, locate every frame facial image characteristic of correspondence point to be identified;

According to the corresponding color histogram of facial image to be identified, determine the crucial facial image in facial image to be identified;

Determine the key feature points in the crucial facial image character pair point of every frame;

Determined crucial facial image is carried out to geometrical normalization pre-service and image irradiation pre-service, to reduce image geometry feature and the illumination impact on crucial facial image;

According to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determine face recognition result corresponding to described video flowing to be identified;

Wherein,

Described according to the corresponding color histogram of facial image to be identified, determine the crucial facial image in facial image to be identified, be specially:

The facial image to be identified that settles the standard, and calculate the color histogram of the corresponding human face region of described standard facial image to be identified;

Calculate the color histogram of the corresponding human face region of every frame facial image to be identified;

Calculate the Euclidean distance of the color histogram that color histogram that current facial image to be identified is corresponding is corresponding with standard facial image to be identified;

Judge whether described Euclidean distance is less than predeterminable range threshold value, if so, described current facial image to be identified is defined as to crucial facial image; If not, described facial image to be identified is defined as to non-key facial image;

Describedly determine face recognition result corresponding to described video flowing to be identified according to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, be specially:

Utilize wavelet transformation to determine the corresponding critical eigenvalue of each key feature points;

Calculate in the crucial facial image of every frame the similarity of each faceform's character pair point in each key feature points and face database;

The similarity of each key feature points and same faceform's character pair point in crucial every frame facial image is weighted, faceform corresponding maximum weighted value is defined as to the alternate recognition results of described crucial facial image;

Add up crucial face amount of images corresponding to each alternative face recognition result, and by quantity at most and be defined as higher than the alternative face recognition result of predetermined number threshold value the face recognition result that video flowing to be identified is corresponding.

2. method according to claim 1, is characterized in that, described every two field picture in described video flowing to be identified is carried out to face detection, to determine facial image to be identified, is specially:

Every two field picture in described video flowing to be identified is carried out to gray processing and histogram equalization processing;

The face that every two field picture after treatment is carried out to multiple dimensioned many features detects;

The corresponding testing result of each yardstick of every two field picture is carried out to many features and merge processing;

Many features corresponding every two field picture are merged to result and carry out multiple dimensioned merging processing;

The multiple dimensioned merging result that meets default face yardstick threshold value is defined as to facial image to be identified.

3. method according to claim 1, is characterized in that, described by crest or trough corresponding to producing under different integral projection modes, locates every frame facial image characteristic of correspondence point to be identified, is specially:

Adopt adaptive threshold method to carry out Enhancement contrast processing to facial image to be identified;

Utilize vertical integral projection method and horizontal integral projection method to carry out positioning feature point processing to facial image to be identified, to locate described facial image characteristic of correspondence point to be identified.

4. method according to claim 1, is characterized in that, described determined crucial facial image is carried out to geometrical normalization pre-service and image irradiation pre-service, is specially:

Described crucial facial image is carried out to translation, rotation and filtering processing, to realize geometrical normalization processing;

Crucial facial image through above-mentioned processing is carried out to histogram equalization processing, to realize illumination pretreatment.

5. the face identification system based on video flowing, is characterized in that, comprising:

Video reception module, the video flowing to be identified gathering for receiver, video collecting device;

Face detection module, for the every two field picture of described video flowing to be identified is carried out to face detection, to determine facial image to be identified;

Positioning feature point module, for by crest or trough corresponding to producing under different integral projection modes, locates every frame facial image characteristic of correspondence point to be identified;

Key frame determination module, for according to the corresponding color histogram of facial image to be identified, determines the crucial facial image in facial image to be identified;

Key point determination module, for determining the key feature points of the crucial facial image character pair point of every frame;

Pretreatment module, for determined crucial facial image is carried out to geometrical normalization pre-service and image irradiation pre-service, to reduce image geometry feature and the illumination impact on crucial facial image;

Result determination module, according to the weighting result of the similarity of each faceform's character pair point in key feature points and face database in the crucial facial image of every frame, determines face recognition result corresponding to described video flowing to be identified;

Wherein,

Described key frame determination module comprises:

Standard unit, for the facial image to be identified that settles the standard, and calculates the color histogram of the corresponding human face region of described standard facial image to be identified;

Color histogram determining unit, for calculating the color histogram of the corresponding human face region of every frame facial image to be identified;

Histogram difference computing unit, for calculating the Euclidean distance of the color histogram that color histogram that current facial image to be identified is corresponding is corresponding with standard facial image to be identified;

Key frame determining unit, for judging whether described Euclidean distance is less than predeterminable range threshold value, if so, is defined as crucial facial image by described current facial image to be identified; If not, described facial image to be identified is defined as to non-key facial image;

Described result determination module comprises:

Eigenwert computing unit, for utilizing wavelet transformation to determine the corresponding critical eigenvalue of each key feature points;

Similarity calculated, calculates in the crucial facial image of every frame the similarity of each faceform's character pair point in each key feature points and face database;

Alternative result determining unit, is weighted the similarity of each key feature points and same faceform's character pair point in crucial every frame facial image, faceform corresponding maximum weighted value is defined as to the alternate recognition results of described crucial facial image;

Recognition result determining unit, for adding up crucial face amount of images corresponding to each alternative face recognition result, and by quantity at most and be defined as higher than the alternative face recognition result of predetermined number threshold value the face recognition result that video flowing to be identified is corresponding.

6. system according to claim 5, is characterized in that, described face detection module comprises:

Equilibrium treatment unit, carries out gray processing and histogram equalization processing to every two field picture in described video flowing to be identified;

Face detecting unit, detects for the face that every two field picture after treatment is carried out to multiple dimensioned many features;

Many features merge cells, merges processing for the corresponding testing result of each yardstick of every two field picture being carried out to many features;

Multiple dimensioned merge cells, carries out multiple dimensioned merging processing for many features corresponding every two field picture are merged to result;

Result determining unit, for being defined as facial image to be identified by the multiple dimensioned merging result that meets default face yardstick threshold value.

7. system according to claim 5, is characterized in that, described positioning feature point module comprises:

Enhancement contrast unit, for adopting adaptive threshold method to carry out Enhancement contrast processing to facial image to be identified;

Positioning feature point unit, utilizes vertical integral projection method and horizontal integral projection method to carry out positioning feature point processing to facial image to be identified, to locate described facial image characteristic of correspondence point to be identified.

8. system according to claim 5, is characterized in that, described pretreatment module comprises:

Geometrical normalization unit, for carrying out translation, rotation and filtering processing to described crucial facial image, to realize geometrical normalization processing;

Illumination pretreatment unit, for the crucial facial image through above-mentioned processing is carried out to histogram equalization processing, to realize illumination pretreatment.

Images