CN101369315A - Human face detection method - Google Patents
Human face detection method Download PDFInfo
- Publication number
- CN101369315A CN101369315A CNA2007100450041A CN200710045004A CN101369315A CN 101369315 A CN101369315 A CN 101369315A CN A2007100450041 A CNA2007100450041 A CN A2007100450041A CN 200710045004 A CN200710045004 A CN 200710045004A CN 101369315 A CN101369315 A CN 101369315A
- Authority
- CN
- China
- Prior art keywords
- face
- people
- eigenwert
- image
- sample
- 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
Abstract
Provided is a face detecting method, first through training positive negative samples, a cascade classifier is obtained to detect whether each subarea corresponding to each eigenvalue does not belong to face component in the image according to the eigenvalue of the image, then the face image to be detected is converted to a face grey chart, which is furthermore converted to a face integral image, then the face integral image is divided into a plurality of sub-integral domains, and then the eigenvalue in the corresponding sub-domain of each sub-integral domain is computed according to the obtained cascade classifier, and based on the computed eigenvalue, the cascade classifier is adopted to detect step by step whether each sub-integral domain does not belong to the face component, to eliminate the sub-integral domains which do not belong to the face component, finally, face repeated are is processed combinedly to determine position and size of the face according to the judgement result. Due to floating point and fixed point operation adopted in the detecting process, detecting speed of the face is effectively advanced, meanwhile appropriative memory is reduced.
Description
Technical field
The present invention relates to a kind of method for detecting human face.
Background technology
At present, people's face detects and has obtained widespread use in the fields such as man-machine interface, video monitoring and content-based retrieval of a new generation.And along with the development of embedded technology and smart machine, people's face detects the requirement that the field of using has engendered mobile and outwork.Yet the computing power of embedded platform is limited, how to reduce calculated load, minimizing memory access number of times and the minimizing storage space that people's face detects in the algorithm rank, has become the task of top priority thereby obtain a real-time embedded face detection system.
Therefore, how to provide the method for detecting human face that a kind of effective travelling speed is fast, committed memory is little to become the problem that those skilled in the art need to be resolved hurrily in fact.
Summary of the invention
The object of the present invention is to provide a kind of method for detecting human face, with the little requirement in storage area that realizes that its operating load in operational process is low, speed fast and take.
In order to achieve the above object, method for detecting human face provided by the invention, comprise step: 1) adopt a plurality of positive samples and negative sample to carry out sample training in advance, detect in the described image cascade classifier that whether does not belong to people's face ingredient with corresponding mutually each subarea of each eigenwert to obtain each eigenwert according to an image, wherein, each eigenwert can calculate by people's face integrogram; 2) facial image to be detected is converted to people's face gray-scale map that each pixel adopts gray-scale value to represent, and and then described people's face gray-scale map conversion is people's face integrogram; 3) according to the sample-size that sample training adopted described people's face integrogram is divided into a plurality of subproducts subregion; 4) calculate the eigenwert in the corresponding subarea in each subproduct subregion respectively according to the cascade classifier that is obtained, and according to each eigenwert that calculates, adopt described cascade classifier to detect each subproduct subregion step by step and whether do not belong to people's face ingredient, to reject the subproduct subregion that does not belong to people's face ingredient, wherein, in testing process, adopt the computing of floating-point fixed point to come each subproduct subregion eigenwert is separately carried out corresponding computing; 5) according to cascade classifiers at different levels the subregional judged result of not disallowable subproduct is further judged in the described facial image to be detected whether comprise facial image.
Wherein, described positive sample and negative sample are measure-alike, and each of formed cascade classifier grade all is that the correct judgement rate that aligns sample is not less than 99.5% strong classifier, and formed cascade classifier has 8 grades, described subarea is the rectangular area, can be according to formula
Calculate each eigenwert, wherein, featur
jBe j eigenwert, ω
iBe i rectangle r in the rectangular area
iWeighted value, rectsum (r
i) be rectangle r
iEnclose the gray integration of image, N is the rectangle number that comprises in the rectangular area.
In sum, method for detecting human face of the present invention changes fixed-point arithmetic into and image is divided into subregion by training multi-stage cascade sorter, floating-point can realize fast detecting to people's face, and its accuracy of detection height, speed is fast, memory demand is little and corresponding memory access number of times is few, can satisfy the little requirement of portable set memory headroom.
Description of drawings
Fig. 1 is the steps flow chart synoptic diagram of method for detecting human face of the present invention.
The Weak Classifier synoptic diagram that Fig. 2 comprises for the first order sorter that forms in the method for detecting human face of the present invention.
Fig. 3 is that the gray integration of a rectangle in the method for detecting human face of the present invention calculates synoptic diagram.
Embodiment
See also Fig. 1, method for detecting human face of the present invention mainly may further comprise the steps:
The first step: adopt a plurality of positive samples and negative sample to carry out sample training in advance, detect in the described image cascade classifier that whether does not belong to people's face ingredient with corresponding mutually each subarea of each eigenwert to obtain each eigenwert according to an image, wherein, each eigenwert can calculate by people's face integrogram, in the present embodiment, adopt 3000 positive samples in advance, 4000 negative samples are set up training sample database, wherein, described positive sample is the image pattern that comprises facial image, described negative sample is not for comprising the image pattern of facial image, and the size of the positive sample that adopts and negative sample is measure-alike, and for example, the sample training process can be as described below:
A) process of strong classifier training: require each grade strong classifier to hit positive sample rate and must not be lower than 99.5%;
B) set that is input as feature of training process uses each feature that training sample database is classified.In each iteration, according to the judged result of Weak Classifier and the weight distribution of sample, select the Weak Classifier h (n) of the Weak Classifier of an error rate minimum as this iteration generation, upgrade the weights of each training sample afterwards, updating strategy is the weights of the sample that increases h (n) institute mistake branch, like this, in iteration next time, these samples that divided by mistake will give bigger attention, after T iteration of process, common property is given birth to T Weak Classifier, is a strong classifier that classification capacity is stronger after the combination, wherein, each Weak Classifier is an eigenwert;
C) strong classifier is cascaded formation cascade classifier.Should follow the cascade sort principle of " earlier heavy back light " during series connection, be about to be placed on the front by the better simply strong classifier of structure that key character more constitutes.
Because known to those skilled in the art the knowing of process of sample training,, in the present embodiment, generated one 8 grades, comprised the cascade classifier of 102 Weak Classifiers altogether so be not described in detail in this.
Second step: facial image to be detected is converted to people's face gray-scale map that each pixel adopts gray-scale value to represent, and and then the conversion of described people's face gray-scale map is people's face integrogram, usually, because facial image to be detected all is colored yuv format, can be converted into corresponding 8 gray scale forms and form people's face gray-scale map, and be that the ultimate principle of people's face integrogram is by gray-scale map conversion: it is each pixel value sum in the rectangle that forms of diagonal line end points that the value of each pixel in integrogram equals with image origin and this pixel, this known to those skilled in the art knowing is so be not described in detail in this.
The 3rd step: according to the sample-size that sample training adopted described people's face integrogram is divided into a plurality of subproducts subregion, for example, the sample-size that is adopted is 30*30, the corresponding subproduct subregion that people's face integrogram can be divided into a plurality of 30*30.
The 4th step: the eigenwert of calculating the corresponding subarea in each subproduct subregion according to the cascade classifier that is obtained respectively, and according to each eigenwert that calculates, adopt described cascade classifier to detect each subproduct subregion step by step and whether do not belong to people's face ingredient, to reject the subproduct subregion that does not belong to people's face ingredient, wherein, in testing process, adopt the computing of floating-point fixed point to come each subproduct subregion eigenwert is separately carried out corresponding computing, in the present embodiment, the Weak Classifier that comprises according to the first order cascade classifier pairing subarea of eigenwert just at first, calculate the eigenwert in corresponding each subarea in each subproduct subregion respectively, see also Fig. 2, suppose that formed first order cascade classifier has comprised 2 Weak Classifiers in the present embodiment, the rectangular area that Weak Classifier 1 is made up of two rectangles of Bai-Hei, and each rectangle comprises 4 pixels, Weak Classifier 2 is by rectangular area that Bai-Hei-three rectangles of Bai form, and each rectangular area comprises 2 pixels, therefore correspondingly, also calculate the eigenwert in two subareas of correspondence position in each subproduct subregion, it can be according to formula
Calculate each eigenwert, wherein, feature
jBe j eigenwert (j=1 or 2), ω
iBe i rectangle r in the rectangular area
iWeighted value (determining its size) by corresponding Weak Classifier, N is the rectangle number that comprises in the rectangular area, rectsum (r
i) be rectangle r
iEnclose the gray integration of image, see also Fig. 3, the rectangle r that forms by pixel A, B, C and D in the Weak Classifier 1
iGray integration value rectsum (r
i) can be: rectsum (r
i)=ii (C)+ii (A)-ii (B)-ii (D), calculate after two eigenwerts, adopt the computing of floating-point fixed point to come subregional two eigenwerts of each subproduct are carried out further computing to obtain detection threshold again, and the threshold value of itself and first order sorter compared, to reject the subproduct subregion that does not belong to people's face ingredient, and then the pairing subarea of Weak Classifier that comprises according to second level sorter, calculate the eigenwert in each not disallowable subregional corresponding subarea of subproduct respectively, copy the testing process of aforementioned first order sorter to detect equally again, detect so step by step, each subproduct subregion that may include facial image can be stayed, reject each the subproduct subregion that does not belong to people's face ingredient, because it is eigenwert is similar to existing calculating process through the calculating process that computing obtains detection threshold, the different floating-point operations that only are that existing computing is adopted, and the present invention adopts is the fixed point computing, so be not described in detail in this.
The 5th step: according to cascade classifiers at different levels the subregional judged result of not disallowable subproduct is further judged in the described facial image to be detected whether comprise facial image, this determination methods is known to those skilled in the art to be known, and is not described in detail in this.
Method for detecting human face of the present invention is applied in portable embedded platform, for example be applied to based on flush bonding processors such as digital signal processing (DSP), ARM or PowerPC, total correct recognition rata of creating successfully, requiring to align sample in the sample storehouse is not less than under 0.95 the situation, generated one 8 grades, when comprising the cascade classifier of 102 Weak Classifiers altogether, storage space only needs 11K, be 1/10 of existing people's face detection matrix storehouse size, and first order sorter can exclude 60% non-face subimage efficiently.Have, be converted to fixed point by floating-point format, the present invention can obtain 7-10 speed doubly and improve on flush bonding processor.Also have, when on DSP platform (for example TMS320DM642 of TI), adopt method for detecting human face of the present invention, can finish people's face to each two field picture of 352x288 size in 40 milliseconds detects, and not only effective to the front face detection, and it is also fine that the people's face under the multiple attitudes such as low-angle pitching, side direction rotation is detected effect.
In sum, method for detecting human face of the present invention changes fixed-point arithmetic into and image is divided into subregion and can detect fast people's face by training multi-stage cascade sorter, floating-point, its accuracy of detection height, speed is fast, memory demand is little and corresponding memory access number of times is few, thereby can be implemented under the less memory headroom of portable set, carrying out people's face fast detects, simultaneously, also can make things convenient for further recognition of face.
Claims (6)
1. method for detecting human face is characterized in that comprising step:
1) adopt a plurality of positive samples and negative sample to carry out sample training in advance, detect in the described image cascade classifier that whether does not belong to people's face ingredient with corresponding mutually each subarea of each eigenwert to obtain each eigenwert according to an image, wherein, each eigenwert can calculate by people's face integrogram;
2) facial image to be detected is converted to people's face gray-scale map that each pixel adopts gray-scale value to represent, and and then described people's face gray-scale map conversion is people's face integrogram;
3) according to the sample-size that sample training adopted described people's face integrogram is divided into a plurality of subproducts subregion;
4) calculate the eigenwert in the corresponding subarea in each subproduct subregion respectively according to the cascade classifier that is obtained, and according to each eigenwert that calculates, adopt described cascade classifier to detect each subproduct subregion step by step and whether do not belong to people's face ingredient, to reject the subproduct subregion that does not belong to people's face ingredient, wherein, in testing process, adopt the computing of floating-point fixed point to come each subproduct subregion eigenwert is separately carried out corresponding computing;
5) according to cascade classifiers at different levels the subregional judged result of not disallowable subproduct is further judged in the described facial image to be detected whether comprise people's face.
2. method for detecting human face as claimed in claim 1 is characterized in that comprising: described positive sample and negative sample are measure-alike.
3. method for detecting human face as claimed in claim 1 is characterized in that comprising: each of formed cascade classifier grade all is that the correct judgement rate that aligns sample is not less than 99.5% strong classifier.
4. method for detecting human face as claimed in claim 1 is characterized in that: formed cascade classifier has 8 grades.
5. method for detecting human face as claimed in claim 1 is characterized in that: described subarea is the rectangular area.
6. method for detecting human face as claimed in claim 5 is characterised in that: according to formula
Calculate each eigenwert, wherein, feature
jBe j eigenwert, ω
iBe i rectangle r in the rectangular area
iWeighted value, rectsum (r
i) be rectangle r
iEnclose the gray integration of image, N is the rectangle number that comprises in the rectangular area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2007100450041A CN101369315A (en) | 2007-08-17 | 2007-08-17 | Human face detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2007100450041A CN101369315A (en) | 2007-08-17 | 2007-08-17 | Human face detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101369315A true CN101369315A (en) | 2009-02-18 |
Family
ID=40413124
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2007100450041A Pending CN101369315A (en) | 2007-08-17 | 2007-08-17 | Human face detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101369315A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101931749A (en) * | 2009-06-25 | 2010-12-29 | 原相科技股份有限公司 | Shooting parameter adjustment method for face detection and image capturing device for face detection |
| CN102129572A (en) * | 2011-02-25 | 2011-07-20 | 杭州海康威视软件有限公司 | Face detection method and device adopting cascade classifier |
| CN102194202A (en) * | 2010-03-01 | 2011-09-21 | 索尼公司 | Image processing equipment and method |
| CN102201059A (en) * | 2011-05-20 | 2011-09-28 | 北京大学深圳研究生院 | Pedestrian detection method and device |
| CN102779265A (en) * | 2011-05-09 | 2012-11-14 | 北京汉邦高科数字技术股份有限公司 | Fixed-point type human face detection method |
| CN102779267A (en) * | 2011-05-12 | 2012-11-14 | 株式会社理光 | Method and device for detection of specific object region in image |
| CN102799894A (en) * | 2012-06-21 | 2012-11-28 | 上海理工大学 | Detection method for rapidly overlooking pedestrians |
| CN101887514B (en) * | 2009-05-13 | 2013-10-30 | 联咏科技股份有限公司 | Human face detecting device and human face detecting method thereof |
| CN103530646A (en) * | 2012-06-12 | 2014-01-22 | 通用汽车环球科技运作有限责任公司 | Complex-object detection using a cascade of classifiers |
| WO2015154305A1 (en) * | 2014-04-11 | 2015-10-15 | Intel Corporation | Object detection using directional filtering |
| CN105913024A (en) * | 2016-04-12 | 2016-08-31 | 上海交通大学 | Android mobile terminal detecting method based on LAP operator for resisting replay attacks |
| CN106156749A (en) * | 2016-07-25 | 2016-11-23 | 福建星网锐捷安防科技有限公司 | Method for detecting human face based on selective search and device |
| CN106570489A (en) * | 2016-11-10 | 2017-04-19 | 腾讯科技（深圳）有限公司 | Living body determination method and apparatus, and identity authentication method and device |
| CN106651748A (en) * | 2015-10-30 | 2017-05-10 | 华为技术有限公司 | Image processing method and apparatus |
| US9697443B2 (en) | 2014-12-11 | 2017-07-04 | Intel Corporation | Model compression in binary coded image based object detection |
| WO2018068533A1 (en) * | 2016-10-10 | 2018-04-19 | 深圳云天励飞技术有限公司 | Human face detection method and device |
| TWI623842B (en) * | 2013-07-31 | 2018-05-11 | Alibaba Group Services Ltd | Image search and method and device for acquiring image text information |
| CN109583267A (en) * | 2017-09-28 | 2019-04-05 | 京东方科技集团股份有限公司 | Vehicle object detection method, vehicle object detecting device and vehicle |
-
2007
- 2007-08-17 CN CNA2007100450041A patent/CN101369315A/en active Pending
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887514B (en) * | 2009-05-13 | 2013-10-30 | 联咏科技股份有限公司 | Human face detecting device and human face detecting method thereof |
| CN101931749A (en) * | 2009-06-25 | 2010-12-29 | 原相科技股份有限公司 | Shooting parameter adjustment method for face detection and image capturing device for face detection |
| CN101931749B (en) * | 2009-06-25 | 2014-01-08 | 原相科技股份有限公司 | Shooting parameter adjustment method for face detection and image capturing device for face detection |
| CN102194202A (en) * | 2010-03-01 | 2011-09-21 | 索尼公司 | Image processing equipment and method |
| CN102129572A (en) * | 2011-02-25 | 2011-07-20 | 杭州海康威视软件有限公司 | Face detection method and device adopting cascade classifier |
| CN102129572B (en) * | 2011-02-25 | 2013-05-15 | 杭州海康威视数字技术股份有限公司 | Face detection method and device adopting cascade classifier |
| CN102779265B (en) * | 2011-05-09 | 2016-01-27 | 北京汉邦高科数字技术股份有限公司 | A kind of detection method of fixed-point type human face |
| CN102779265A (en) * | 2011-05-09 | 2012-11-14 | 北京汉邦高科数字技术股份有限公司 | Fixed-point type human face detection method |
| CN102779267A (en) * | 2011-05-12 | 2012-11-14 | 株式会社理光 | Method and device for detection of specific object region in image |
| CN102779267B (en) * | 2011-05-12 | 2015-08-12 | 株式会社理光 | The method and apparatus of specific object region in detected image |
| CN102201059A (en) * | 2011-05-20 | 2011-09-28 | 北京大学深圳研究生院 | Pedestrian detection method and device |
| CN103530646A (en) * | 2012-06-12 | 2014-01-22 | 通用汽车环球科技运作有限责任公司 | Complex-object detection using a cascade of classifiers |
| CN103530646B (en) * | 2012-06-12 | 2017-08-15 | 通用汽车环球科技运作有限责任公司 | The complex object cascaded using grader is detected |
| CN102799894A (en) * | 2012-06-21 | 2012-11-28 | 上海理工大学 | Detection method for rapidly overlooking pedestrians |
| CN102799894B (en) * | 2012-06-21 | 2015-02-25 | 上海理工大学 | Detection method for rapidly overlooking pedestrians |
| TWI623842B (en) * | 2013-07-31 | 2018-05-11 | Alibaba Group Services Ltd | Image search and method and device for acquiring image text information |
| WO2015154305A1 (en) * | 2014-04-11 | 2015-10-15 | Intel Corporation | Object detection using directional filtering |
| US10121090B2 (en) | 2014-04-11 | 2018-11-06 | Intel Corporation | Object detection using binary coded images and multi-stage cascade classifiers |
| CN106575364A (en) * | 2014-04-11 | 2017-04-19 | 英特尔公司 | Object detection using directional filtering |
| TWI617996B (en) * | 2014-04-11 | 2018-03-11 | 美商英特爾公司 | Object detection using directional filtering |
| US9697443B2 (en) | 2014-12-11 | 2017-07-04 | Intel Corporation | Model compression in binary coded image based object detection |
| US9940550B2 (en) | 2014-12-11 | 2018-04-10 | Intel Corporation | Model compression in binary coded image based object detection |
| CN106651748A (en) * | 2015-10-30 | 2017-05-10 | 华为技术有限公司 | Image processing method and apparatus |
| CN106651748B (en) * | 2015-10-30 | 2019-10-22 | 华为技术有限公司 | A kind of image processing method and image processing apparatus |
| US10740657B2 (en) | 2015-10-30 | 2020-08-11 | Huawei Technologies Co., Ltd. | Image processing method and image processing apparatus |
| CN105913024B (en) * | 2016-04-12 | 2019-05-24 | 上海交通大学 | The Android mobile phone endpoint detection methods of resistance Replay Attack based on LAP operator |
| CN105913024A (en) * | 2016-04-12 | 2016-08-31 | 上海交通大学 | Android mobile terminal detecting method based on LAP operator for resisting replay attacks |
| CN106156749A (en) * | 2016-07-25 | 2016-11-23 | 福建星网锐捷安防科技有限公司 | Method for detecting human face based on selective search and device |
| WO2018068533A1 (en) * | 2016-10-10 | 2018-04-19 | 深圳云天励飞技术有限公司 | Human face detection method and device |
| CN106570489A (en) * | 2016-11-10 | 2017-04-19 | 腾讯科技（深圳）有限公司 | Living body determination method and apparatus, and identity authentication method and device |
| CN109583267A (en) * | 2017-09-28 | 2019-04-05 | 京东方科技集团股份有限公司 | Vehicle object detection method, vehicle object detecting device and vehicle |
| CN109583267B (en) * | 2017-09-28 | 2022-04-08 | 京东方科技集团股份有限公司 | Vehicle target detection method, vehicle target detection device, and vehicle |
| US11482013B2 (en) | 2017-09-28 | 2022-10-25 | Beijing Boe Technology Development Co., Ltd. | Object tracking method, object tracking apparatus, vehicle having the same, and computer-program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101369315A (en) | Human face detection method | |
| CN110390691B (en) | Ore dimension measuring method based on deep learning and application system | |
| CN110059554B (en) | Multi-branch target detection method based on traffic scene | |
| Li et al. | Automatic detection and classification system of domestic waste via multimodel cascaded convolutional neural network | |
| CN111062413A (en) | Road target detection method and device, electronic equipment and storage medium | |
| CN102496001B (en) | Method of video monitor object automatic detection and system thereof | |
| CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
| WO2019114036A1 (en) | Face detection method and device, computer device, and computer readable storage medium | |
| CN101944174B (en) | Identification method of characters of licence plate | |
| CN106874868A (en) | A kind of method for detecting human face and system based on three-level convolutional neural networks | |
| CN102867183B (en) | Method and device for detecting littered objects of vehicle and intelligent traffic monitoring system | |
| CN107844785A (en) | A kind of method for detecting human face based on size estimation | |
| CN103208008A (en) | Fast adaptation method for traffic video monitoring target detection based on machine vision | |
| CN107909027A (en) | It is a kind of that there is the quick human body target detection method for blocking processing | |
| Medina-Carnicer et al. | Unimodal thresholding for edge detection | |
| CN102147866A (en) | Target identification method based on training Adaboost and support vector machine | |
| CN112200143A (en) | Road disease detection method based on candidate area network and machine vision | |
| CN105930794A (en) | Indoor scene identification method based on cloud computing | |
| CN105225523A (en) | A kind of parking space state detection method and device | |
| US20170091573A1 (en) | Visual attention detector and visual attention detection method | |
| Xiang et al. | Lightweight fully convolutional network for license plate detection | |
| CN105608457B (en) | Gray Histogram square thresholding method | |
| CN101251896B (en) | Object detecting system and method based on multiple classifiers | |
| CN101908214B (en) | Moving object detection method with background reconstruction based on neighborhood correlation | |
| CN102982534A (en) | Canny edge detection dual threshold acquiring method based on chord line tangent method |
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 |
Open date: 20090218 |