CN102096823A - Face detection method based on Gaussian model and minimum mean-square deviation - Google Patents
Face detection method based on Gaussian model and minimum mean-square deviation Download PDFInfo
- Publication number
- CN102096823A CN102096823A CN 201110036818 CN201110036818A CN102096823A CN 102096823 A CN102096823 A CN 102096823A CN 201110036818 CN201110036818 CN 201110036818 CN 201110036818 A CN201110036818 A CN 201110036818A CN 102096823 A CN102096823 A CN 102096823A
- Authority
- CN
- China
- Prior art keywords
- skin
- face
- colour
- image
- ycbcr
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a face detection method based on a Gaussian model and a minimum mean-square deviation, and relates to a face recognition technology. A face detection method based on the Gaussian model and the minimum mean-square deviation under the premise of a complicated background, a side face, a stopper and a plurality of faces. The method comprises the following steps of: building a YCbCr Gaussian model: building the YCbCr Gaussian model for face skin color distribution according to collected skin color sample data, and performing lighting compensation on the image, wherein in the YCbCr, Y is a brightness component, Cb is a blue chroma component, and Cr is a red chroma; performing skin color segmentation on the image by using the built YCbCr Gaussian model and the minimum mean-square deviation; performing binaryzation on a skin color region, and processing a binary image by opening to eliminate a small bridge and discrete points; rejecting detected non-face regions in the similar skin color or skin color according to future knowledge of a face; and finally marking a face position by using a rectangular frame.
Description
Technical field
The present invention relates to a kind of face recognition technology, especially relate to a kind of under the situation of complex background, side, shelter, plurality of human faces the method for quick of colour image human face.
Background technology
People's face detects and to be meant for given image arbitrarily, adopts certain strategy that it is searched for determining wherein whether to exist people's face, if exist then further determine the information such as position, size and attitude of people's face.Along with the enhancing of ecommerce and people's awareness of safety, people are to confirming that personal identification has higher requirement, and the technology of utilizing biological characteristic to discern personal identification also has been endowed very high expectation.And face recognition technology is accepted by people because of people's face obtains than fingerprint iris is easier, becomes the most potential biometric verification of identity means to be measured.At present, people's face detect be widely used in video conference, object-based coding, use vision system (as safety inspection, guarding management etc.), Web search, three-dimensional face are synthetic, video frequency searching, CBIR etc.
The detection of people's face is one and has very big complicacy, challenge, its main difficult point can reduce: whether have people's face in (1), the image: how to judge whether there is people's face in the image, how to distinguish the non-face image of people's face and similar people's face, how under complicated background, to extract the feature of people's face etc. accurately; (2), detect people's face of different expression form: people's face may appear in the image with different visual angles, people in piece image is bold and also may differs huge for a short time, also may be blocked by other object, such as existing under a plurality of people's the scene, block just easier appearance of situation, cause some to be used to detect and the face characteristic that needs to extract is invisible; (3), the influence of image: have noise in the image, the influence of factors such as brightness during owing to imaging, contrast makes not fogging clear, and people's face and background difference are little, bring certain difficulty to detection; For coloured image,, the method for detecting human face based on colour is brought very big influence with making if image has the appearance of situations such as colour cast; (4), the factor of people's face self: people's face has different attitude (front, 45 or side etc.), make that the variation of people's face pattern is very big, some attitude even can shelter from face's organ (for example eyes and nose etc.), special face's structural information is such as the glasses of face and beard etc.; The difference expression of face is also very big to the influence of people's face pattern.
Above-mentioned factor all is to solve people's face detection problem to have caused difficulty, if can find some relevant algorithms and can reach in application process in real time, will detect with application system such as tracking and give security for successfully constructing the people's face with actual application value.The method for detecting human face of Ti Chuing can be divided into four classes in recent years: based on knowledge method, based on template matching method, based on the method for face shaping with based on the method for feature.
1., template matches at present the main stream approach that detects of people's face has:: at first set up some face templates of storage as standard, these templates can be people's face or independent eyes, nose, mouth or rubber-like template, utilize some algorithms to calculate the similarity degree or the correlativity of each zone to be measured and standard form then, judge with this whether a certain zone is people's face or individual features; Relative merits: the algorithm based on template matches has been obtained effect preferably in the utilization at the single face in front, detect for people's face of complex background, set up one and can well distinguish background interference to represent the template of each species diversity people face again simultaneously be the key point of problem.The template that fine differentiation background interference is arranged, be difficult to represent various face templates, and the template that can represent various people's faces can not be distinguished background well, it also is the common problem that exists during people's face detects, accuracy rate and loss are a pair of contradiction all the time, and the computation complexity of template matches is higher, and the live effect of location is also bad, and is also bad to the detection effect of side.2., neural network: rely on statistical study and machine learning techniques, construct nerve network system, use a large amount of people's faces and non-face sample that system is trained, allow system learn the class sigma-t of two class sample complexity automatically as sorter; Relative merits: have higher accuracy, fault-tolerance, robustness, avoided complex features extraction work, obtain the recessiveness expression that additive method is difficult to realize by learning process about the recognition of face rule, though neural net method is very feasible in theory, but need to carry out a large amount of sample trainings, relative velocity is slower, and testing result is subjected to the influence of sample also bigger.3., Face Detection: utilize the chromatic information of digital picture, be partitioned into possible people's face candidate regions earlier, differentiate at each candidate regions then, determine whether candidate regions is people's face; Relative merits: the key of colour of skin localization method is to seek a suitable colour system coordinate system, it can make the variation of the colour of skin concentrate on certain two dimension, setting threshold is to determine the colour of skin thus, make that the non-colour of skin in this threshold range is few more good more, detection speed is very fast relatively, substantially can accomplish the influence that real-time detection is not changeed by the rotation of people's face or people's side, loss is very low.
Chinese patent CN101667245 discloses a kind of based on the method for detecting human face of supporting the vector cascading novel detection classifiers, what mainly solve is excessive long problem detection time that causes of computation complexity in people's face testing process, and its testing process is: the pre-service of training sample set and extraction sample characteristics; Utilize and support the novel detection algorithm of vector that the training sample that extracts feature is trained, and obtain sorter model; According to just examining with false drop rate to come the optimized Algorithm parameter and select suitable three suitable sub-classifiers to be cascaded into a strong classifier in the detection of existing test set; The gray scale picture is carried out the detection of people's face and mark with this cascade device of assigning to by force.This invention has the fast advantage of detection speed, and the people's face that can be used in machine learning and the pattern-recognition category detects.
Chinese patent CN101957909A discloses a kind of method for detecting human face based on the DSP platform.The method is by to dynamic video cut-away view picture, and utilizes image processing means, based on the model integrity attribute, considers the topological relation between people's face integral body and the each several part, extracts and utilizes key message to seek unique point, detects thereby finish whole man's face.The system hardware part is made up of following three parts: the front end image acquisition partly adopts the CCD camera of band infrared photography function; The BF533 processor is partly adopted in information processing; The rear end is connected TV and directly shows people's face testing result with the DSP platform.Does software section adopt Visual? the DSP++ development environment is programmed and is realized data collection, processing, output control.
Summary of the invention
The objective of the invention is to provides a kind of method for detecting human face based on Gauss model and Minimum Mean Square Error under the situation of complex background, side, shelter, plurality of human faces at the defective of existing algorithm existence and the difficult point of people's face detection.
The present invention includes following steps:
1) set up the YCbCr Gauss model: set up the YCbCr Gauss model that face complexion distributes according to the colour of skin sample data of gathering, image is carried out light compensation, among the described YCbCr, Y is a luminance component, and Cb is the chroma blue component, and Cr is the red color component;
In step 1), the described YCbCr Gauss model of setting up the face complexion distribution according to the colour of skin sample data of gathering, be according to the colour of skin sample data of gathering, count the average m and the variance c of sample, the YCbCr Gauss model that utilizes the m and the c parameter fitting colour of skin to distribute; Described image is carried out light compensation, be to exist color error ratio in the entire image in order to offset, the brightness of all pixels in the entire image is arranged from high to low, get preceding 5% pixel, if the number of these pixels abundant (for example greater than 100), just with their brightness as " reference white " (Reference White), R, G, the B component value that is about to their color all is adjusted into maximum 255.
2) the YCbCr Gauss model and the Minimum Mean Square Error of utilization foundation carry out skin color segmentation to image;
In step 2) in, YCbCr Gauss model and Minimum Mean Square Error that described utilization is set up can be the concrete steps that image carries out skin color segmentation: according to the colour of skin sample data of gathering, set up the KL space, determine the threshold value of the colour of skin in the KL space; Utilize the YCbCr Gauss model to calculate the likelihood score image of the colour of skin, utilize the adaptive threshold method that image is cut apart then; To utilize the threshold value in KL space to obtain area of skin color at last and combine, thereby obtain colour of skin mixture model by Minimum Mean Square Error and YCbCr Gauss model with the area of skin color that the YCbCr Gauss model obtains.
3) area of skin color is carried out binaryzation, and the utilization opening operation is handled removal foot bridge or discrete point to bianry image;
In step 3), described binaryzation to area of skin color, can use the colour of skin mixture model that combines based on YCbCr Gauss model and Minimum Mean Square Error, if a pixel promptly satisfies the colour of skin threshold value in KL space, also belong to simultaneously the pixel that YCbCr Gauss model and adaptive threshold split, judge that then this point is a colour of skin point, otherwise be background dot; Described opening operation can carry out opening operation with the binary image after the skin color segmentation, promptly corrodes the process of after expansion earlier; Corroding method is, takes the initial point of structural element S and the point on the X to contrast singly, if the institute on the S has a few all in the scope of X, and the then point of the initial point correspondence of S reservation, otherwise this point is removed; The method that expands is, takes the initial point of structural element S and the point on the X to contrast singly, if having a point to drop in the scope of X on the S, then the point of the initial point correspondence of S just keeps, otherwise this point is removed; Handle through opening operation, can remove isolated point, burr and foot bridge (point that promptly is communicated with two zones).
4) reject the non-face zone of the detected class colour of skin or the colour of skin according to the priori of people's face, use rectangle frame labelling human face position at last.
In step 4), described priori according to people's face is rejected the non-face zone of the detected class colour of skin or the colour of skin, can reject the non-face zone of the class colour of skin or the colour of skin according to the priori of people's face in image; The priori of described people's face in image can comprise the ratio that people's face should occupy (be pixel what), the length breadth ratio of people's face and the closeness (occupation rate of people's face pixel) of human face region etc. in image; Described labelling human face, can reject the non-face zone of the class colour of skin or the colour of skin according to priori after, remaining zone is exactly so-called human face region, irises out with rectangle frame.
The described colour of skin mixture model that combines based on YCbCr Gauss model and Minimum Mean Square Error, be meant the colour of skin data of elder generation according to the not agnate facial image under the different conditions such as colour of skin sample, indoor and outdoor illumination deficiency under colour of skin sample, the shade under the natural lighting of gathering, draw the colour of skin at KL spatial distributions figure, thereby determine the threshold value of face complexion in the KL space; And count the average m and the variance c of sample, utilize the YCbCr Gauss model of the m and c parameter fitting colour of skin distribution, calculate the likelihood score image of the colour of skin, utilize the adaptive threshold method that image is cut apart then; At last the threshold value in KL space is obtained area of skin color that area of skin color and YCbCr Gauss model obtain with, thereby obtain the area of skin color of people's face.
The non-face module of the described removal class colour of skin or the colour of skin, be meant through after the processing of opening operation, next each area of skin color is carried out mark, and calculate each regional length and width and the shared pixel of pixel and whole zone, the non-face zone that utilizes the priori of people's face to reject some the class colours of skin or the colour of skin then, wherein the priori of people's face comprises: 1. the minimum number of pixels of people's face is not less than 600, wherein widely is not less than 20, longly is not less than 30; 2. the depth-width ratio ratio of people's face boundary rectangle, 0.8＜ratio＜2; 3. the number of pixels of people's face accounts for boundary rectangle and is not less than 45%; Reject non-face zone according to above 3 conditions.
The present invention be a kind of under the situation of complex background, side, shelter, plurality of human faces the method for quick of colour image human face, when realizing a face detection system, at first should collect abundant face complexion sample, (YCbCr wherein Y is meant luminance component to set up the YCbCr Gauss model that face complexion distributes according to a large amount of colour of skin samples of gathering, Cb refers to the chroma blue component, and Cr refers to the red color component) and definite colour of skin in KL spatial distributions scope; Next according to Fig. 1 flow process, image is carried out light compensation, YCbCr Gauss model and Minimum Mean Square Error that utilization is set up carry out skin color segmentation fast to image, then area of skin color is carried out binaryzation, and the utilization opening operation is handled bianry image, remove some foot bridges or discrete point, then according to the priori (size of area of skin color of people's face, length breadth ratio, filling rate) the non-face zone of the rejecting class colour of skin or the colour of skin uses rectangle (boundary rectangle of people's face) frame to iris out people's face position accurately at last.
This shows, outstanding advantage of the present invention is: at the defective of existing algorithm existence and the difficult point of people's face detection, the present invention proposes a kind of colour of skin mixture model that combines with the YCbCr Gauss model based on Minimum Mean Square Error, accurately area of skin color is cut apart fast, then the area of skin color that splits is carried out morphology and handle (being opening operation), after the morphology processing, the priori of utilization people face is screened the area of skin color of handling, propose the non-face zone of some the class colours of skin or the colour of skin, mark people's face position at last.
In addition, utilization of the present invention is based on Gauss's complexion model of YCbCr, and successful separates monochrome information with color information, and has higher Face Detection rate; By Minimum Mean Square Error, can reduce the mean square deviation of each component, Gauss model and Minimum Mean Square Error are combined, just further dwindled the scope that the colour of skin is filtered.
Simultaneously, the used Face Detection of the present invention mainly is to utilize the cluster of the colour of skin at color space, and meets two-dimentional Gaussian distribution, by the colour of skin sample that collects, the statistics colour of skin is in the cluster areas of color space, and sets up Gauss model with the method for mathematics and obtain a function.The present invention complex background, side, shelter, plurality of human faces etc. can locate the position of people's face under the situation fast, the part difficult point of present people's face detection and the part defective of existing algorithm have been solved, strong robustness, and have higher detection speed and verification and measurement ratio, therefore be with a wide range of applications.
Experimental result shows that the segmentation effect that this regional model merges Gauss model likelihood figure is better than only adopting any method, and this is not see reported method so far.
Description of drawings
Fig. 1 is the face detection system process flow diagram.In Fig. 1, flow process is for opening image, light compensation, and KL conversion, based on the YCbCr Gaussian transformation, binaryzation, opening operation (corrosion after expansion earlier) goes to false face zone, the labelling human face.
Fig. 2 is colour of skin distribution Gauss model after the match.In Fig. 2, horizontal ordinate is respectively C
bComponent and C
rComponent, ordinate are probability distribution.
Fig. 3 is the erosion operation example.Fig. 3 is the XY coordinate, and in Fig. 3, the left side is X, and the centre is S, and the right side is X-S.
Fig. 4 is the dilation operation example.Fig. 4 is the XY coordinate, and in Fig. 4, the left side is X, and the centre is S, and the right side is X+S.
Fig. 5 is structural element S of the present invention.Fig. 5 is the XY coordinate.
Fig. 6 is for to remove the process flow diagram in non-face zone based on priori.In Fig. 6, flow process combines with the KL conversion for the Gauss's complexion model based on YCbCr, candidate zone (binary map), zone marker, add up the shared pixel of each regional length and width and pixel and whole zone, calculate area of skin color size, the length breadth ratio of zoning, the filling rate of zoning, the labelling human face.
Embodiment
A kind of under the situation of complex background, side, shelter, plurality of human faces the method for quick of colour image human face, when realizing a face detection system, at first should collect abundant face complexion sample, (YCbCr wherein Y is meant luminance component to set up the YCbCr Gauss model that face complexion distributes according to a large amount of colour of skin samples of gathering, Cb refers to the chroma blue component, and Cr refers to the red color component) and definite colour of skin in KL spatial distributions scope; Next according to Fig. 1 flow process, image is carried out light compensation, YCbCr Gauss model and Minimum Mean Square Error that utilization is set up carry out skin color segmentation fast to image, then area of skin color is carried out binaryzation, and the utilization opening operation is handled bianry image, remove some foot bridges or discrete point, then according to the priori (size of area of skin color of people's face, length breadth ratio, filling rate) the non-face zone of the rejecting class colour of skin or the colour of skin uses rectangle (boundary rectangle of people's face) frame to iris out people's face position accurately at last.
A, skin color modeling:
Earlier, draw the colour of skin at KL spatial distributions figure, thereby determine the threshold value of face complexion in the KL space according to the colour of skin data of the not agnate facial image under the natural lighting of gathering, under the shade, under the different condition such as indoor and outdoor illumination deficiency; And count the average m and the variance c of sample, utilize the YCbCr Gauss model of the m and c parameter fitting colour of skin distribution, calculate the likelihood score image of the colour of skin, utilize the adaptive threshold method that image is cut apart then; At last the threshold value in KL space is obtained area of skin color and combine, thereby obtain the colour of skin mixture model that combines by YCbCr Gauss model and Minimum Mean Square Error with the area of skin color that the YCbCr Gauss model obtains.
The colour of skin mixture model based on YCbCr Gauss model and Minimum Mean Square Error that the present invention adopts is:
(1) Minimum Mean Square Error
Pixel in the image is transformed in the KL space according to (1-1), when pixel in the image satisfies (1-2) in the KL space, then this pixel is for belonging to area of skin color, otherwise do not belong to area of skin color, and its Chinese style (1-2) is to be determined by the sample data of gathering.Conversion formula from rgb space to the KL space:
The threshold value of the colour of skin is:
110.2＜K1＜376.3，-61.3＜K2＜32.9，-18.8＜K3＜19.5 (1-2)
(2) YCbCr Gauss model
1. colour of skin likelihood score image
Can be write the pixel in the sample as following formula from rgb space to the YCbCr space conversion:
C according to the colour of skin sample that collects
r, C
b, (m, c), concrete computing formula is as follows to set up a two-dimentional Gauss model G
C＝E[(x-m)(x-m)
T] (1-7)
X is the chroma vector of each pixel; First component is Cr; Second component is Cb; M and C are respectively average and the variances that comes out.Through experiment statistics, average and variance are respectively:
m＝(156.5599，117.4361)
T
Simulate colour of skin distribution Gauss model according to above-mentioned parameter, calculating formula of similarity is as follows:
P(C
b，C
r)＝exp[-0.5(x-m)
TC
-1(x-m)] (1-8)
According to the complexion model of having set up, calculate the possibility size of all pixel colors and the colour of skin in the facial image, i.e. similarity size, its span is [0,1].Wherein: x=[C
r, C
b] be the vector of testing image pixel in the Cb-Cr space.After calculating finishes, to each P[C that is tried to achieve
r, C
b] value carries out normalized, is about to each P[C
r, C
b] be worth divided by P[C maximum in this image
r, C
b] value, so just make the value of similarity of each pixel between [0,1], and make area of skin color brightness more outstanding.This value is big more, and the possibility that expression belongs to the colour of skin is also big more, otherwise more little.
2. threshold value chooses
The present invention mainly by the adaptive threshold method, further is converted into binary map with the similarity image, and wherein 0,1 represents non-area of skin color and area of skin color respectively.The method of Cai Yonging is to allow threshold value reduce since 0.65 in the present invention, each minimizing 0.1, till 0.05, and belong to the variation of skin pixel quantity when noting each changes of threshold, find out then and belong to skin pixel number change that threshold value hour as the optimization threshold value, as obtain increasing minimumly reducing to 0.35 quantity of skin pixel when interval from 0.45, the threshold value after then optimizing is 0.40.Then think this point for skin pixel and put 1 when the similarity of this pixel is higher than the optimization threshold value, otherwise then put 0 if be lower than this threshold value.
Comprehensively 1. it is as follows, 2. can to draw the skin color modeling concrete grammar:
I, calculating maximum likelihood degree are obtained the colour of skin likelihood score of each pixel correspondence according to formula (1-8), and are obtained the maximum colour of skin likelihood score of entire image;
II, calculate colour of skin likelihood score image, with the colour of skin likelihood score of each pixel divided by the resulting value of maximum colour of skin likelihood score as this gray values of pixel points, obtain colour of skin likelihood score image, the likelihood value of each pixel has characterized the probability that this pixel belongs to skin;
III, skin color segmentation adopt the adaptive threshold method that image is cut apart.At first setting threshold is 0.65, reduce up to 0.05 with 0.1 interval at every turn, belong to the variation of skin pixel quantity when noting each changes of threshold, find out that threshold value that the pixel quantity that belongs to the colour of skin changes hour then as optimal threshold, with the likelihood image binaryzation;
IV, the bianry image that obtains are done dot product with the result that Minimum Mean Square Error is cut apart again, to eliminate certain class colour of skin influence.
B, interpreting blueprints and light compensation
Process flow diagram according to Fig. 1, at first read in view data, then the brightness of all pixels in the entire image is arranged from high to low, get preceding 5% pixel, and with their brightness as " reference white " (Reference White), R, G, the B component value that is about to their color all is adjusted into maximum 255.
C, skin color segmentation and binaryzation
1. image is obtained the colour of skin likelihood score of each pixel correspondence, and is obtained the maximum colour of skin likelihood score of entire image through after the pre-service according to formula (1-8); 2. the colour of skin likelihood score of using each pixel obtains colour of skin likelihood score image divided by the likelihood value of the resulting value of maximum colour of skin likelihood score as this pixel; 3. adopt the adaptive threshold method that image is cut apart.At first setting threshold is 0.65, reduce up to 0.05 with 0.1 interval at every turn, belong to the variation of skin pixel quantity when noting each changes of threshold, find out that threshold value that the pixel quantity that belongs to the colour of skin changes hour then as optimal threshold, the similarity of pixel is higher than the optimization threshold value and thinks that then this point is skin pixel and put 1 in image, otherwise then puts 0 if be lower than this threshold value; 4. when pixel in the image satisfied (1-2) in the KL space, then this pixel did not then put 0 for belonging to area of skin color and putting 1 otherwise do not belong to area of skin color; 5. satisfying the colour of skin point that KL space and YCbCr Gauss model are determined simultaneously, be set to 255, other pixels are changed to 0.
D, opening operation are handled
The structural element S that the present invention adopts be Fig. 5, takes the initial point of S and the point on the area of skin color to contrast singly, if on the S have a few all in the scope of area of skin color the then point of the initial point correspondence of S reservation, otherwise this point is removed; Take the initial point of S and the point of the area of skin color after the corrosion to contrast singly, if there is a point to drop in the scope of X on the S, then the point of the initial point correspondence of S is a colour of skin point just, otherwise is not;
E, based on the non-face module of the rejecting of priori
Behind Gauss's complexion model and Minimum Mean Square Error processing of image process based on YCbCr, obtain bianry image, and then bianry image is carried out opening operation handle removal discrete point and foot bridge; Next mark is carried out in each zone, and calculates each regional length and width and the shared pixel of pixel and whole zone, the non-face zone that utilizes the priori of people's face to reject some the class colours of skin or the colour of skin then:
(1) the minimum pixel count of identification people face
If the width of target area is greater than W (the present invention gets 20), height then keeps this zone, otherwise deletes this zone greater than H (the present invention gets 30);
(2) depth-width ratio of people's face
If the high wide ratio ratio of the boundary rectangle of target area not in the threshold range (0.8～2.5) of regulation the time, then deletes this zone.In general, the depth-width ratio of people's face is approximately 1, considers that people's attitude is different, for anti-leak-stopping choosing, stipulates that the following of this ratio is limited to 0.8, and all of ratio＜0.8 zones to be checked will be by filtering.On the other hand, people's face depth-width ratio also should be stipulated a upper limit.This is because have certain situation in practice, contain people's face in the zone to be checked, but the depth-width ratio of image has exceeded the scope of the normal depth-width ratio of people's face.Because neck and following skin area thereof expose to some extent, the to be checked regional depth-width ratio that obtains by skin color segmentation has just exceeded normal scope in this case as the people.Therefore must consider this special circumstances, provide a wideer depth-width ratio upper limit 2.5.By the use of regular shape, just can get rid of some irregular but and the approaching object of skin color, also can get rid of simultaneously other non-face zones of some human bodies, as the four limbs of bending etc.;
(3) occupation rate rule
If the closeness C of target area then keeps this zone, otherwise deletes this zone greater than 45%.
If area of skin color is discontented with any one that is enough to three conditions, then delete this zone, if all satisfy, then keep this zone.
F, labelling human face
When all area of skin color after treatment, remaining zone is exactly a human face region, and each piece human face region all has a mark mark.Next executor's face labeling algorithm: 1. measure the value of the high order end and the low order end in every zone to be marked, the value of the value-high order end of the wide=low order end in zone then to be marked (every); 2. measure the epimere in every zone to be marked, then bottom=epimere-2* is wide; 3. use rectangle (boundary rectangle of people's face) the frame human face region that draws.
Utilize above step, detected everyone face position in can output image.
In skin color modeling of the present invention, regional model is a kind of better simply complexion model, is convenient to understand and utilization, but determines the threshold value difficulty.Simple Gauss model opposed area model can better represent that the colour of skin distributes, and the Face Detection rate is higher, and model parameter is easy to calculate, but speed is slower than regional model.In order better to detect the colour of skin, this patent adopts the skin color segmentation method that combines based on Gauss model and Minimum Mean Square Error (this method adopts the KL conversion).Face Detection is carried out in simple use KL conversion, though can obtain effect preferably, still seems undesirable.Main cause is that the result vector Y of simple KL conversion is subjected to the influence of input vector X very big, because transformation matrix itself is constructed according to input vector X, in essence, the groundwork that the KL conversion is finished is three components that 3 components of script linear dependence converted to linear independence, thereby realizes redistributing of colouring information.Three components that the colour of skin information of people's face is re-assigned to equally all comprise the colour of skin information of people's face, and this brings difficulty for follow-up Threshold Segmentation.Although the mean square deviation of each component is all enough little, is applied to Threshold Segmentation and still seems excessive.Equally, the YCbCr space is not to aim at people's face to detect designed color space, and it carries out therefore simple use people's face and detect effect and may also can not get desirable effect.Based on traditional YCbCr space and KL conversion deficiency separately, combine them formation YCbCr-KL color space.So construct new image binaryzation function F=M ∩ N, color cut apart the colour of skin binary map that obtains be defined as follows:
Morphology is handled: morphologic computing is based on corrosion and these two kinds of fundamental operations of expanding, what the present invention adopted is opening operation, promptly corrode the process of after expansion earlier, former figure is through behind the opening operation, can remove isolated point, burr and foot bridge (point that promptly is communicated with two zones), eliminate wisp, level and smooth larger object border, while and its area of not obvious change; Corrosion: the effect in mathematical morphology is a frontier point of eliminating object, makes the process of border to internal contraction, can remove the object less than structural element.Choose the structural element of different sizes like this, just can remove the object of different sizes.As between two objects tiny connection being arranged, can be separately by corrosion with two objects; Expand: the effect in mathematical morphology is just in time opposite with the effect of corrosion, and it is that binaryzation object boundary point is expanded, and all background dots that will contact with object merge in this object, makes the process of border to the outside expansion.If the distance between two objects is closer, then dilation operation may be communicated to two objects together, expands to filling up after the image segmentation cavity in the object of great use.
Reject non-face: owing to may exist the non-face zone of the class colour of skin or the colour of skin (such as arm in the image, thigh etc.), therefore must verify the area of skin color that screens, if this area of skin color can't satisfy in three conditions of people's face priori any one, this area of skin color is rejected at once.People's face priori: 1. the number of pixels of people's face must be greater than threshold k (when people's face number of pixels reaches certain value, could divide and differentiate people's face, generally get 600); 2. the depth-width ratio ratio of the boundary rectangle of people's face must satisfy certain limit, as not satisfying, rejects this zone; 3. people's face is generally ellipse, and the number of pixels of whole people's face must account for the r% of the number of pixels that people's face boundary rectangle comprised, otherwise rejects this zone.
The present invention consists of the following components: light compensation module, skin color modeling, morphology processing module, based on the non-face module of the rejecting of priori, labelling human face module.
1, light compensation module
Colour of skin isochrome multimedia message breath often is subjected to the influence of the factors such as color error ratio of light source colour, image capture device in the picture, depart from essential color on the whole and move to a certain direction, be our usually said color error ratio, so must carry out pre-service to image; Exist color error ratio in order to offset in this entire image, the brightness of all pixels in the entire image is arranged from high to low, get preceding 5% pixel, if the number of these pixels is abundant (for example, greater than 100), just with their brightness as " reference white " (Reference White), R, G, the B component value that is about to their color all is adjusted into maximum 255.
2, skin color modeling
At first carry out the collection of colour of skin sample, comprise different facial images and not agnate people's face in the scene such as indoor, outdoor; Extract the normal area of skin color data of people's face then, draw the colour of skin, determine the threshold value of the colour of skin in the KL space at KL spatial distributions figure; According to a large amount of colour of skin sample datas of gathering, count the average m and the variance c of sample, the YCbCr Gauss model that utilizes the m and the c parameter fitting colour of skin to distribute; Utilize the YCbCr Gauss model to calculate the likelihood score image of the colour of skin, utilize the adaptive threshold method that image is cut apart then; At last the threshold value in KL space is obtained area of skin color and combine, thereby obtain colour of skin mixture model by YCbCr Gauss model and Minimum Mean Square Error with the area of skin color that the YCbCr Gauss model obtains.
3, morphology processing module (opening operation)
Image is carried out morphology to be handled, mainly be in order to remove isolated point, burr and foot bridge (point that promptly is communicated with two zones), to eliminate wisp, reducing influence to next step processing, the present invention carries out opening operation to the area of skin color that extracts and handles, and promptly corrodes after expansion earlier;
(1) corrosion
Corroding method is, takes the initial point of S and the point on the X to contrast singly, if the institute on the S has a few all in the scope of X, and the then point of the initial point correspondence of S reservation, otherwise this point is removed; Fig. 3, and lacks than the point that X comprises still in the scope of original X through the X Θ S after the corrosion as can be seen, just as the X one deck that has been corroded.
(2) expand
The method that expands is, takes the initial point of S and the point on the X to contrast singly, if there is a point to drop in the scope of X on the S, then the point of the initial point correspondence of S is an image just; Fig. 4 as can be seen, after expand
All scopes that comprise X, just as X expanded one the circle.
4, based on the non-face module of the rejecting of priori
Behind Gauss's complexion model and Minimum Mean Square Error processing of image process based on YCbCr, obtain bianry image, and then bianry image is carried out opening operation handle removal discrete point and foot bridge; Next mark is carried out in each zone, and calculate the shared pixel of each regional length and width and whole zone.The priori of people's face:
(1) the minimum pixel count of identification people face
If the width of target area is less than W, height is then deleted this zone (in general, people's face should occupy certain proportion in image) less than H;
(2) length breadth ratio of people's face
If the high wide ratio ratio of the boundary rectangle of target area not in the threshold range (0.8～2.5) of regulation the time, then deletes this zone.
(3) occupation rate rule
If the closeness C of target area, then deletes this zone (it is oval that human face region generally is approximately) less than a%.
According to above priori, just can reject the non-face zone of the class colour of skin or the colour of skin.
5, labelling human face module
After all area of skin color process pre-service, remaining zone is exactly a human face region, and each piece human face region all has a mark mark.Next executor's face labeling algorithm: 1. measure the value of the high order end and the low order end in every zone to be marked, the value of the value-high order end of the wide=low order end in zone then to be marked (every); 2. measure the epimere in every zone to be marked, then bottom=epimere+2* is wide; 3. use rectangle (boundary rectangle of people's face) the frame human face region that draws.
In order to verify validity of the present invention, can detect at wear dark glasses, plurality of human faces, side, background complexity and people's face backlight, can draw from test result, people's face detection algorithm of the present invention's design is for shelter, the robustness of aspects such as side face, plurality of human faces, light influence is very strong, and can reach the effect of real-time processing.
Colour of skin distribution Gauss model is referring to Fig. 2 after the match.Remove the process flow diagram in non-face zone referring to Fig. 6 based on priori.
Claims (9)
1. based on the method for detecting human face of Gauss model and Minimum Mean Square Error, it is characterized in that may further comprise the steps:
1) set up the YCbCr Gauss model: set up the YCbCr Gauss model that face complexion distributes according to the colour of skin sample data of gathering, image is carried out light compensation, among the described YCbCr, Y is a luminance component, and Cb is the chroma blue component, and Cr is the red color component;
2) the YCbCr Gauss model and the Minimum Mean Square Error of utilization foundation carry out skin color segmentation to image;
3) area of skin color is carried out binaryzation, and the utilization opening operation is handled removal foot bridge or discrete point to bianry image;
4) reject the non-face zone of the detected class colour of skin or the colour of skin according to the priori of people's face, use rectangle frame labelling human face position at last.
2. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1, it is characterized in that in step 1), the described YCbCr Gauss model of setting up the face complexion distribution according to the colour of skin sample data of gathering, be according to the colour of skin sample data of gathering, count the average m and the variance c of sample, the YCbCr Gauss model that utilizes the m and the c parameter fitting colour of skin to distribute.
3. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1, it is characterized in that in step 1), described image is carried out light compensation, be to exist color error ratio in the entire image in order to offset, the brightness of all pixels in the entire image is arranged from high to low, get preceding 5% pixel, if the number of these pixels is greater than 100, just with their brightness as " reference white ", R, G, the B component value that is about to their color all is adjusted into maximum 255.
4. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1, it is characterized in that in step 2) in, YCbCr Gauss model and Minimum Mean Square Error that described utilization is set up to the concrete steps that image carries out skin color segmentation are: according to the colour of skin sample data of gathering, set up the KL space, determine the threshold value of the colour of skin in the KL space; Utilize the YCbCr Gauss model to calculate the likelihood score image of the colour of skin, utilize the adaptive threshold method that image is cut apart then; To utilize the threshold value in KL space to obtain area of skin color at last and combine, thereby obtain colour of skin mixture model by Minimum Mean Square Error and YCbCr Gauss model with the area of skin color that the YCbCr Gauss model obtains.
5. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1, it is characterized in that in step 3), described binaryzation to area of skin color, the colour of skin mixture model that use combines based on YCbCr Gauss model and Minimum Mean Square Error, if a pixel promptly satisfies the colour of skin threshold value in KL space, also belong to simultaneously the pixel that YCbCr Gauss model and adaptive threshold split, judge that then this point is a colour of skin point, otherwise be background dot.
6. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1 is characterized in that in step 3), and described opening operation is that the binary image after the skin color segmentation is carried out opening operation, promptly corrodes the process of after expansion earlier; Corroding method is, takes the initial point of structural element S and the point on the X to contrast singly, if the institute on the S has a few all in the scope of X, and the then point of the initial point correspondence of S reservation, otherwise this point is removed; The method that expands is, takes the initial point of structural element S and the point on the X to contrast singly, if having a point to drop in the scope of X on the S, then the point of the initial point correspondence of S just keeps, otherwise this point is removed; Handle through opening operation, can remove isolated point, burr and foot bridge.
7. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1, it is characterized in that in step 4), described priori according to people's face is rejected the non-face zone of the detected class colour of skin or the colour of skin, be according to the priori of people's face in image, reject the non-face zone of the class colour of skin or the colour of skin.
8. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 7 is characterized in that the priori of described people's face in the image ratio that should occupy that comprises people's face, the length breadth ratio of people's face and the closeness of human face region in image.
9. the method for detecting human face based on Gauss model and Minimum Mean Square Error as claimed in claim 1, it is characterized in that in step 4), described labelling human face, be after rejecting the non-face zone of the class colour of skin or the colour of skin according to priori, remaining zone is exactly so-called human face region, irises out with rectangle frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110036818 CN102096823A (en) | 2011-02-12 | 2011-02-12 | Face detection method based on Gaussian model and minimum mean-square deviation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110036818 CN102096823A (en) | 2011-02-12 | 2011-02-12 | Face detection method based on Gaussian model and minimum mean-square deviation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102096823A true CN102096823A (en) | 2011-06-15 |
Family
ID=44129911
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110036818 Pending CN102096823A (en) | 2011-02-12 | 2011-02-12 | Face detection method based on Gaussian model and minimum mean-square deviation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102096823A (en) |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102254327A (en) * | 2011-07-29 | 2011-11-23 | 西南交通大学 | Method for automatically segmenting face in digital photo |
| CN102324025A (en) * | 2011-09-06 | 2012-01-18 | 北京航空航天大学 | Human face detection and tracking method based on Gaussian skin color model and feature analysis |
| CN102510437A (en) * | 2011-10-25 | 2012-06-20 | 重庆大学 | Method for detecting background of video image based on distribution of red, green and blue (RGB) components |
| CN103247150A (en) * | 2013-05-15 | 2013-08-14 | 苏州福丰科技有限公司 | Fatigue driving preventing system |
| CN103632132A (en) * | 2012-12-11 | 2014-03-12 | 广西工学院 | Face detection and recognition method based on skin color segmentation and template matching |
| CN104318558A (en) * | 2014-10-17 | 2015-01-28 | 浙江大学 | Multi-information fusion based gesture segmentation method under complex scenarios |
| CN104331690A (en) * | 2014-11-17 | 2015-02-04 | 成都品果科技有限公司 | Skin color face detection method and system based on single picture |
| CN104732206A (en) * | 2015-03-12 | 2015-06-24 | 苏州阔地网络科技有限公司 | Human face detecting method and device |
| CN105894020A (en) * | 2016-03-30 | 2016-08-24 | 重庆大学 | Specific target candidate box generating method based on gauss model |
| CN105960801A (en) * | 2014-02-03 | 2016-09-21 | 谷歌公司 | Enhancing video conferences |
| WO2016154781A1 (en) * | 2015-03-27 | 2016-10-06 | Intel Corporation | Low-cost face recognition using gaussian receptive field features |
| CN106274393A (en) * | 2016-08-29 | 2017-01-04 | 北京汽车研究总院有限公司 | The control method of automobile sun-shade-curtain, device and automobile |
| CN106326862A (en) * | 2016-08-25 | 2017-01-11 | 广州御银自动柜员机技术有限公司 | Multi-face pickup device |
| CN106611429A (en) * | 2015-10-26 | 2017-05-03 | 腾讯科技（深圳）有限公司 | Method and device for detecting skin area |
| CN107370981A (en) * | 2016-05-13 | 2017-11-21 | 中兴通讯股份有限公司 | The information cuing method and device of personnel participating in the meeting in a kind of video conference |
| CN107390573A (en) * | 2017-06-28 | 2017-11-24 | 长安大学 | Intelligent wheelchair system and control method based on gesture control |
| CN107633252A (en) * | 2017-09-19 | 2018-01-26 | 广州市百果园信息技术有限公司 | Skin color detection method, device and storage medium |
| CN108171135A (en) * | 2017-12-21 | 2018-06-15 | 深圳云天励飞技术有限公司 | Method for detecting human face, device and computer readable storage medium |
| CN108985249A (en) * | 2018-07-26 | 2018-12-11 | 京东方科技集团股份有限公司 | Method for detecting human face, device, electronic equipment and storage medium |
| CN109165600A (en) * | 2018-08-27 | 2019-01-08 | 浙江大丰实业股份有限公司 | Stage performance personnel's intelligent search platform |
| CN109214363A (en) * | 2018-10-23 | 2019-01-15 | 广东电网有限责任公司 | A kind of substation's worker's face identification method based on YCbCr and connected component analysis |
| CN110163927A (en) * | 2019-05-17 | 2019-08-23 | 温州大学 | A kind of single image neural network based restains method |
| CN110188640A (en) * | 2019-05-20 | 2019-08-30 | 北京百度网讯科技有限公司 | Face identification method, device, server and computer-readable medium |
| WO2019223582A1 (en) * | 2018-05-24 | 2019-11-28 | Beijing Didi Infinity Technology And Development Co., Ltd. | Target detection method and system |
| CN110706237A (en) * | 2019-09-06 | 2020-01-17 | 上海衡道医学病理诊断中心有限公司 | Diaminobenzidine separation and evaluation method based on YCbCr color space |
| CN112541860A (en) * | 2019-09-23 | 2021-03-23 | 深圳开阳电子股份有限公司 | Skin color beautifying correction method and device |
| CN113888543A (en) * | 2021-08-20 | 2022-01-04 | 北京达佳互联信息技术有限公司 | Skin color segmentation method and device, electronic equipment and storage medium |
| CN113947568A (en) * | 2021-09-26 | 2022-01-18 | 北京达佳互联信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN116363736A (en) * | 2023-05-31 | 2023-06-30 | 山东农业工程学院 | Big data user information acquisition method based on digitalization |
| CN116844198A (en) * | 2023-05-24 | 2023-10-03 | 北京优创新港科技股份有限公司 | Method and system for detecting face attack |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101377813A (en) * | 2008-09-24 | 2009-03-04 | 上海大学 | Method for real time tracking individual human face in complicated scene |
-
2011
- 2011-02-12 CN CN 201110036818 patent/CN102096823A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101377813A (en) * | 2008-09-24 | 2009-03-04 | 上海大学 | Method for real time tracking individual human face in complicated scene |
Non-Patent Citations (1)
| Title |
|---|
| 《中国电子学会第十五届信息论学术年会暨第一届全国网络编码学术年会论文集（上册）》 20081231 李作勇等 多彩色空间联合分割的人脸检测 , * |
Cited By (50)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102254327A (en) * | 2011-07-29 | 2011-11-23 | 西南交通大学 | Method for automatically segmenting face in digital photo |
| CN102324025A (en) * | 2011-09-06 | 2012-01-18 | 北京航空航天大学 | Human face detection and tracking method based on Gaussian skin color model and feature analysis |
| CN102324025B (en) * | 2011-09-06 | 2013-03-20 | 北京航空航天大学 | Human face detection and tracking method based on Gaussian skin color model and feature analysis |
| CN102510437A (en) * | 2011-10-25 | 2012-06-20 | 重庆大学 | Method for detecting background of video image based on distribution of red, green and blue (RGB) components |
| CN103632132B (en) * | 2012-12-11 | 2017-02-15 | 广西科技大学 | Face detection and recognition method based on skin color segmentation and template matching |
| CN103632132A (en) * | 2012-12-11 | 2014-03-12 | 广西工学院 | Face detection and recognition method based on skin color segmentation and template matching |
| CN103247150A (en) * | 2013-05-15 | 2013-08-14 | 苏州福丰科技有限公司 | Fatigue driving preventing system |
| CN105960801A (en) * | 2014-02-03 | 2016-09-21 | 谷歌公司 | Enhancing video conferences |
| CN105960801B (en) * | 2014-02-03 | 2020-02-07 | 谷歌有限责任公司 | Enhancing video conferencing |
| CN104318558A (en) * | 2014-10-17 | 2015-01-28 | 浙江大学 | Multi-information fusion based gesture segmentation method under complex scenarios |
| CN104318558B (en) * | 2014-10-17 | 2017-06-23 | 浙江大学 | Hand Gesture Segmentation method based on Multi-information acquisition under complex scene |
| CN104331690B (en) * | 2014-11-17 | 2017-08-29 | 成都品果科技有限公司 | A kind of colour of skin method for detecting human face and system based on single image |
| CN104331690A (en) * | 2014-11-17 | 2015-02-04 | 成都品果科技有限公司 | Skin color face detection method and system based on single picture |
| CN104732206A (en) * | 2015-03-12 | 2015-06-24 | 苏州阔地网络科技有限公司 | Human face detecting method and device |
| WO2016154781A1 (en) * | 2015-03-27 | 2016-10-06 | Intel Corporation | Low-cost face recognition using gaussian receptive field features |
| US10872230B2 (en) | 2015-03-27 | 2020-12-22 | Intel Corporation | Low-cost face recognition using Gaussian receptive field features |
| CN106611429B (en) * | 2015-10-26 | 2019-02-05 | 腾讯科技（深圳）有限公司 | Detect the method for skin area and the device of detection skin area |
| CN106611429A (en) * | 2015-10-26 | 2017-05-03 | 腾讯科技（深圳）有限公司 | Method and device for detecting skin area |
| US10783353B2 (en) | 2015-10-26 | 2020-09-22 | Tencent Technology (Shenzhen) Company Limited | Method for detecting skin region and apparatus for detecting skin region |
| CN105894020B (en) * | 2016-03-30 | 2019-04-12 | 重庆大学 | Specific objective candidate frame generation method based on Gauss model |
| CN105894020A (en) * | 2016-03-30 | 2016-08-24 | 重庆大学 | Specific target candidate box generating method based on gauss model |
| CN107370981A (en) * | 2016-05-13 | 2017-11-21 | 中兴通讯股份有限公司 | The information cuing method and device of personnel participating in the meeting in a kind of video conference |
| CN106326862A (en) * | 2016-08-25 | 2017-01-11 | 广州御银自动柜员机技术有限公司 | Multi-face pickup device |
| CN106274393A (en) * | 2016-08-29 | 2017-01-04 | 北京汽车研究总院有限公司 | The control method of automobile sun-shade-curtain, device and automobile |
| CN107390573A (en) * | 2017-06-28 | 2017-11-24 | 长安大学 | Intelligent wheelchair system and control method based on gesture control |
| CN107390573B (en) * | 2017-06-28 | 2020-05-29 | 长安大学 | Intelligent wheelchair system based on gesture control and control method |
| US11080894B2 (en) | 2017-09-19 | 2021-08-03 | Bigo Technology Pte. Ltd. | Skin color detection method, skin color detection apparatus, and storage medium |
| CN107633252A (en) * | 2017-09-19 | 2018-01-26 | 广州市百果园信息技术有限公司 | Skin color detection method, device and storage medium |
| CN107633252B (en) * | 2017-09-19 | 2020-04-21 | 广州市百果园信息技术有限公司 | Skin color detection method, device and storage medium |
| CN108171135A (en) * | 2017-12-21 | 2018-06-15 | 深圳云天励飞技术有限公司 | Method for detecting human face, device and computer readable storage medium |
| WO2019223582A1 (en) * | 2018-05-24 | 2019-11-28 | Beijing Didi Infinity Technology And Development Co., Ltd. | Target detection method and system |
| CN108985249A (en) * | 2018-07-26 | 2018-12-11 | 京东方科技集团股份有限公司 | Method for detecting human face, device, electronic equipment and storage medium |
| CN109165600B (en) * | 2018-08-27 | 2021-11-26 | 浙江大丰实业股份有限公司 | Intelligent search platform for stage performance personnel |
| CN109165600A (en) * | 2018-08-27 | 2019-01-08 | 浙江大丰实业股份有限公司 | Stage performance personnel's intelligent search platform |
| CN109214363A (en) * | 2018-10-23 | 2019-01-15 | 广东电网有限责任公司 | A kind of substation's worker's face identification method based on YCbCr and connected component analysis |
| CN110163927A (en) * | 2019-05-17 | 2019-08-23 | 温州大学 | A kind of single image neural network based restains method |
| CN110163927B (en) * | 2019-05-17 | 2023-04-07 | 温州大学 | Single image re-coloring method based on neural network |
| CN110188640A (en) * | 2019-05-20 | 2019-08-30 | 北京百度网讯科技有限公司 | Face identification method, device, server and computer-readable medium |
| CN110188640B (en) * | 2019-05-20 | 2022-02-25 | 北京百度网讯科技有限公司 | Face recognition method, face recognition device, server and computer readable medium |
| CN110706237A (en) * | 2019-09-06 | 2020-01-17 | 上海衡道医学病理诊断中心有限公司 | Diaminobenzidine separation and evaluation method based on YCbCr color space |
| CN110706237B (en) * | 2019-09-06 | 2023-06-06 | 上海衡道医学病理诊断中心有限公司 | Diamino benzidine separation and evaluation method based on YCbCr color space |
| CN112541860A (en) * | 2019-09-23 | 2021-03-23 | 深圳开阳电子股份有限公司 | Skin color beautifying correction method and device |
| CN113888543A (en) * | 2021-08-20 | 2022-01-04 | 北京达佳互联信息技术有限公司 | Skin color segmentation method and device, electronic equipment and storage medium |
| CN113888543B (en) * | 2021-08-20 | 2024-03-19 | 北京达佳互联信息技术有限公司 | Skin color segmentation method and device, electronic equipment and storage medium |
| CN113947568A (en) * | 2021-09-26 | 2022-01-18 | 北京达佳互联信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN113947568B (en) * | 2021-09-26 | 2024-03-29 | 北京达佳互联信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN116844198A (en) * | 2023-05-24 | 2023-10-03 | 北京优创新港科技股份有限公司 | Method and system for detecting face attack |
| CN116844198B (en) * | 2023-05-24 | 2024-03-19 | 北京优创新港科技股份有限公司 | Method and system for detecting face attack |
| CN116363736A (en) * | 2023-05-31 | 2023-06-30 | 山东农业工程学院 | Big data user information acquisition method based on digitalization |
| CN116363736B (en) * | 2023-05-31 | 2023-08-18 | 山东农业工程学院 | Big data user information acquisition method based on digitalization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102096823A (en) | Face detection method based on Gaussian model and minimum mean-square deviation | |
| CN101630363B (en) | Rapid detection method of face in color image under complex background | |
| CN105809138B (en) | A kind of road warning markers detection and recognition methods based on piecemeal identification | |
| CN102214291B (en) | Method for quickly and accurately detecting and tracking human face based on video sequence | |
| CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
| CN100361138C (en) | Method and system of real time detecting and continuous tracing human face in video frequency sequence | |
| CN105160317B (en) | One kind being based on area dividing pedestrian gender identification method | |
| CN102799901B (en) | Method for multi-angle face detection | |
| WO2018072233A1 (en) | Method and system for vehicle tag detection and recognition based on selective search algorithm | |
| CN102194108B (en) | Smile face expression recognition method based on clustering linear discriminant analysis of feature selection | |
| CN103473571B (en) | Human detection method | |
| CN101840509B (en) | Measuring method for eye-observation visual angle and device thereof | |
| CN105046206B (en) | Based on the pedestrian detection method and device for moving prior information in video | |
| CN104951773A (en) | Real-time face recognizing and monitoring system | |
| CN105046219A (en) | Face identification system | |
| CN101383001A (en) | Quick and precise front human face discriminating method | |
| CN105138954A (en) | Image automatic screening, query and identification system | |
| CN109558825A (en) | A kind of pupil center's localization method based on digital video image processing | |
| CN101763504A (en) | Human head identification method under complex scene | |
| CN101923645A (en) | Iris splitting method suitable for low-quality iris image in complex application context | |
| CN101853397A (en) | Bionic human face detection method based on human visual characteristics | |
| CN103020614B (en) | Based on the human motion identification method that space-time interest points detects | |
| CN106503748A (en) | A kind of based on S SIFT features and the vehicle targets of SVM training aids | |
| El Maghraby et al. | Detect and analyze face parts information using Viola-Jones and geometric approaches | |
| CN111860291A (en) | Multi-mode pedestrian identity recognition method and system based on pedestrian appearance and gait information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20110615 |