CN105512638A - Fused featured-based face detection and alignment method - Google Patents
Fused featured-based face detection and alignment method Download PDFInfo
- Publication number
- CN105512638A CN105512638A CN201510983142.9A CN201510983142A CN105512638A CN 105512638 A CN105512638 A CN 105512638A CN 201510983142 A CN201510983142 A CN 201510983142A CN 105512638 A CN105512638 A CN 105512638A
- Authority
- CN
- China
- Prior art keywords
- face
- image
- feature
- characteristic
- input
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a fused featured-based face detection and alignment method. The method includes the following four steps that: multi-scale pyramid images are generated by an input image, the images of various of scales are scanned with a size-fixed window and a fixed step length, so that a plurality of candidate face windows can be generated; the candidate face windows generated by a scanning window module are transferred to a face and non-face classifier as input, the classifier outputs a face window; a face image outputted in the last step is transferred to a face feature point regression device as input, and the face feature point regression device outputs face feature point positions in the face image, such as eyebrows, eyes, nose, mouth and the like; and rotation and scaling processing is performed on a face according to the face image outputted by a face feature point detection module and face feature point positions corresponding to the face image, and an aligned face image is outputted. With the fused featured-based face detection and alignment method of the invention adopted, automatic detection and automatic alignment of a face in an image can be realized. The fused featured-based face detection and alignment method has the advantages of high speed, high accuracy and the like, and is conductive to improving the accuracy rate of face verification and face recognition technologies.
Description
Technical field
The present invention relates to technical field of computer vision, be specifically related to a kind of Face datection based on traditional characteristic and convolutional neural networks feature and alignment schemes.
Background technology
Along with the development of computer science, man-machine interaction has become more and more valued technology.Start to be applied in industry member as the recognition of face of computer vision field and face verification technology, recognition of face and verification technique are the hot research problems of computer vision field always in the past few decades.And Face datection is a vital step in recognition of face with aliging.
First, the more common method for detecting human face of current use is the method for detecting human face based on Haar-like characteristic sum AdaBoost technology, and it is by extracting the face classification device of the features training of engineer based on AdaBoost technology.But due to the low level abstract characteristics that Haar-like feature is a kind of engineer, do not have complete face information, so cause the sorter accuracy rate of training out not high.
Secondly, the man face characteristic point positioning method that current use is more common is adaptive shape model method (ASM, adaptiveshapemodel).This method does not possess very strong robustness for off-note point and attitudes vibration, is therefore difficult to obtain human face characteristic point accurately, and this will directly affect face alignment effect, cause the degradation of recognition of face performance further.
Again, although higher than classic method in accuracy rate based on the method for convolutional neural networks, convolutional neural networks calculated amount is large, process single picture length consuming time, is difficult to reach detect face in real time and the requirement of alignment.So need a kind of new Face datection of exploitation and alignment schemes, the method based on traditional characteristic and the methods combining based on convolutional neural networks can be got up.
In order to solve the problem, the invention provides a kind of Face datection based on traditional characteristic and convolutional neural networks feature and alignment schemes, the method, to comprise arbitrarily the image of face for input, can detect and the face that aligns rapidly and accurately.
Summary of the invention
The technical matters that the present invention solves is: overcome existing based on traditional characteristic with based on the Face datection of convolutional neural networks and the deficiency of alignment schemes, provide a kind of Face datection based on fusion feature and alignment schemes.
The technical solution used in the present invention is: a kind of Face datection based on fusion feature and alignment schemes, and schematic flow sheet of the present invention as shown in Figure 7, comprises following four steps:
Step (1), first input picture is amplified to 5 times to original image size, then with 0.7937 for coefficient is multiplied by the picture size after amplification, generate new sized image, repeat this process until picture size size is less than specific threshold (as 110), after generating multi-Scale Pyramid image according to above-mentioned steps, then scan the image of different scale with fixed size window and fixed step size, generate multiple candidate face window;
Step (2), the traditional characteristic (localbinarypattern characteristic sum Haar-like feature) then extracting image and convolutional neural networks feature, train a non-face sorter of face; The non-face sorter of the face that training in advance is good is two sorters in essence, and input fixed size image, exports 0 or 1, namely represents whether input picture is face; Candidate face window step (1) generated successively passes to the non-face sorter of the good face of training in advance as input, and sorter exports face window;
Step (3), the traditional characteristic (localbinarypattern characteristic sum Haar-like feature) then extracting facial image and convolutional neural networks feature, train a human face characteristic point to return device; The facial image exported with step (2) is that input passes to the good human face characteristic point recurrence device of training in advance, and human face characteristic point recurrence device exports the human face characteristic point position (eyebrow, canthus, nose, the corners of the mouth etc.) in this facial image;
The human face characteristic point position of step (4), the facial image finally exported according to step (3) and correspondence thereof, choose two fixing human face characteristic points (as canthus, left and right) as fixed point, Rotation and Zoom process is carried out to facial image, export the facial image after alignment, in the facial image after alignment, the position of selected fixed characteristic points is by constant.
Further, the multi-scale image pyramid model described in step (1), with 0.7937 for zoom factor generates multi-scale image.
Further, the non-face sorter of face described in step (2) is that input training obtains by extracting the feature of image, and the feature of image to be merged mutually with convolutional neural networks feature by traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and forms.
Further, it be feature by extracting image is that input training obtains that human face characteristic point described in step (3) returns device, and the feature of image to be merged mutually with convolutional neural networks feature by traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and forms.
Further, carrying out Rotation and Zoom described in step (4) to facial image, is process based on the guarantee face left and right tail of the eye 2 is constant.
Principle of the present invention is:
The invention provides a kind of Face datection based on fusion feature and alignment schemes, the method, to comprise arbitrarily the image of face for input, can detect and the face that aligns rapidly and accurately.This method comprises four steps: first generate multi-Scale Pyramid image by input picture, then with fixed size window and fixed step size scanning different scale images, generates multiple candidate face window; Then successively the candidate face window of scanning window CMOS macro cell is passed to the non-face sorter of face as input, sorter exports face window; Then the facial image exported with previous step returns device for input passes to human face characteristic point, and human face characteristic point recurrence device exports the human face characteristic point position (eyebrow, canthus, nose, the corners of the mouth etc.) in this facial image; Finally according to the facial image of facial feature points detection module output and the human face characteristic point position of correspondence thereof, Rotation and Zoom process is carried out to face, exports the facial image after alignment.The present invention can realize automatic detection to face in image and automatic aligning, has the advantages that speed is fast, accuracy rate is high, contributes to the accuracy rate improving face verification and face recognition technology.
Content of the present invention mainly includes following four steps:
(1) scanning window step: first input picture is amplified to 5 times to original image size, then with 0.7937 for coefficient is multiplied by the picture size after amplification, generate new sized image, repeat this process until picture size size is less than specific threshold (as 110), after generating multi-Scale Pyramid image according to above-mentioned steps, then scan the image of different scale with fixed size window and fixed step size, generate multiple candidate face window; Be no matter the facial image of which kind of size, the window obtained after above process can include all people's face in image.
(2) the non-face classifying step of face: this module needs the non-face sorter of what a face of training in advance, extract traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and the convolutional neural networks feature of image, train a non-face sorter of face; This sorter is two sorters in essence, and input fixed size image, exports 0 or 1, namely represents whether input picture is face; Then successively the candidate face window of scanning window CMOS macro cell is passed to the non-face sorter of face as input, sorter exports face window.The face classification device obtained is trained both to overcome the not high shortcoming of traditional characteristic method accuracy rate by after traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and convolutional neural networks Fusion Features, drastically increase counting yield again, thus the object quick and precisely detecting face can be reached.
(3) facial feature points detection step: this module needs training in advance human face characteristic point to return device equally, extract traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and the convolutional neural networks feature of facial image, train a human face characteristic point to return device; This recurrence device receives fixed size facial image, exports the human face characteristic point position (eyebrow, canthus, nose, the corners of the mouth etc.) in this facial image.
(4) face alignment step: the input of this module is the facial image of facial feature points detection module output and the human face characteristic point position of correspondence thereof, exports the facial image after alignment.Face alignment module is alignd according to two fixing human face characteristic points (as canthus, left and right), Rotation and Zoom process is carried out to facial image, export the facial image after alignment, in the facial image after alignment, the position of fixed characteristic points selected by previous step is by constant.
The present invention's advantage is compared with prior art:
1, the face classification device training method of the present invention's proposition, first LBP characteristic sum Haar-like feature is extracted, then convolutional neural networks feature is extracted, a face sorter is trained in conjunction with these three kinds of characteristic bindings. train the face classification device obtained both to overcome the not high shortcoming of traditional characteristic method accuracy rate by after traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and convolutional neural networks Fusion Features, drastically increase counting yield again, thus the object quick and precisely detecting face can be reached.
2, the human face characteristic point that the present invention proposes returns device training method, first extracts LBP characteristic sum Haar-like feature, then extracts convolutional neural networks feature, train a face to return device in conjunction with these three kinds of characteristic bindings..The human face characteristic point having merged traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and convolutional neural networks feature returns device and significantly improves accuracy rate when not reducing counting yield.
3, the present invention propose Face datection and alignment schemes to illumination, attitude, block stronger robustness, have higher lifting to the performance of follow-up recognition of face and face verification.
Accompanying drawing explanation
Fig. 1 is based on process flow diagram of the present invention;
Fig. 2 is LBP feature templates schematic diagram;
Fig. 3 is Haar-like feature templates schematic diagram;
Fig. 4 is the face classification device design sketch that the present invention trains;
Fig. 5 is that human face characteristic point returns device result schematic diagram;
Fig. 6 is face alignment result schematic diagram;
Fig. 7 is schematic flow sheet of the present invention.
Embodiment
Fig. 1 gives based on traditional characteristic and the Face datection of convolutional neural networks feature and the overall process flow of alignment schemes, further illustrates the present invention below in conjunction with other the drawings and the specific embodiments.
The invention provides the Face datection based on traditional characteristic and convolutional neural networks feature and alignment schemes, key step is described below:
1, scanning window step
First input picture is amplified to 5 times to original image size, then with 0.7937 for coefficient is multiplied by the picture size after amplification, generate new sized image, repeat this process until picture size size is less than specific threshold (as 110), after generating multi-Scale Pyramid image according to above-mentioned steps, then scan the image of different scale with fixed size window and fixed step size, generate multiple candidate face window.
2, the non-face classifying step of face
Extract traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and the convolutional neural networks feature of image, train a non-face sorter of face; The non-face sorter of the face that training in advance is good is two sorters in essence, and input fixed size image, exports 0 or 1, namely represents whether input picture is face; Candidate face window step (1) generated successively passes to the non-face sorter of the good face of training in advance as input, and sorter exports face window.
1) traditional characteristic is extracted
First the present invention extracts the LBP characteristic sum Haar-like feature of facial image according to existing LBP Feature Extraction Technology and Haar-like Feature Extraction Technology, then merge this two category feature extracted, concrete steps are as follows:
The annular neighborhood formwork calculation of LBP Feature Extraction Technology according to Fig. 2 goes out the LBP code of central pixel point, then the LBP code of entire image is divided into several zonules, get the statistic histogram in each region, the histogram in all regions is coupled together the LBP textural characteristics just obtaining view picture figure.The feature templates of Haar-like Feature Extraction Technology according to Fig. 3 calculates Haar-like eigenwert, utilizes the integrogram technology of Haar-like Feature Extraction Technology can the Haar-like feature of rapid extraction entire image.After extracting LBP characteristic sum Haar-like feature, two kinds of features are tied, then utilize principal component analysis (PCA) technology to be tieed up the Feature Dimension Reduction to 200 that is tied, this dimension be 200 feature be the fusion of the traditional characteristic extracted.
2) convolutional neural networks feature is extracted
The present invention as training dataset, trained a convolutional neural networks sorter using 200,000 facial images (from 10574 different people), but using layer second from the bottom as characteristic layer.Namely input picture gets layer value second from the bottom as feature after the convolutional neural networks sorter that this trains.
3) training classifier
After having extracted traditional characteristic and convolutional neural networks feature according to above-mentioned definition, splice these two kinds of features successively as final fusion feature, input training two sorters are characterized as with this, this two sorter, with the fusion feature of image for input, export 0 or 1, namely represent whether input picture is face, the effect of two sorters face in detected image that final training obtains as shown in Figure 4, is irised out position with circle and is represented the face detected.
3, facial feature points detection step
First facial feature points detection module extracts traditional characteristic (localbinarypattern characteristic sum Haar-like feature) and the convolutional neural networks feature of facial image, then trains a human face characteristic point to return device; This recurrence device receives fixed size facial image, export human face characteristic point position in this facial image (eyebrow, canthus, nose, the corners of the mouth etc., as shown in Figure 5).
4, face alignment step
The input of this module is the facial image of facial feature points detection module output and the human face characteristic point position of correspondence thereof, exports the facial image after alignment.Face alignment module is alignd according to two fixing human face characteristic points (as canthus, left and right), Rotation and Zoom process is carried out to facial image, export the facial image after alignment, in the facial image after alignment, the position of fixed characteristic points selected by previous step is by constant (as shown in Figure 6).
The technology contents that the present invention does not elaborate belongs to the known technology of those skilled in the art.
Although be described the illustrative embodiment of the present invention above; so that the technician of this technology neck understands the present invention; but should be clear; the invention is not restricted to the scope of embodiment; to those skilled in the art; as long as various change to limit and in the spirit and scope of the present invention determined, these changes are apparent, and all innovation and creation utilizing the present invention to conceive are all at the row of protection in appended claim.
Claims (5)
1., based on Face datection and the alignment schemes of fusion feature, it is characterized in that comprising following four steps:
Step (1), first input picture is amplified to 5 times to original image size, then with 0.7937 for coefficient is multiplied by the picture size after amplification, generate new sized image, repeat this process until picture size size is less than specific threshold, after generating multi-Scale Pyramid image according to above-mentioned steps, then scan the image of different scale with fixed size window and fixed step size, generate multiple candidate face window;
Step (2), the traditional characteristic then extracting image and convolutional neural networks feature, train a non-face sorter of face, and described traditional characteristic refers to localbinarypattern characteristic sum Haar-like feature; The non-face sorter of the face that training in advance is good is two sorters in essence, and input fixed size image, exports 0 or 1, namely represents whether input picture is face; Candidate face window step (1) generated successively passes to the non-face sorter of the good face of training in advance as input, and sorter exports facial image;
Step (3), the traditional characteristic then extracting facial image and convolutional neural networks feature, train a human face characteristic point to return device; The facial image exported with step (2) is that input passes to the good human face characteristic point recurrence device of training in advance, and human face characteristic point recurrence device exports the human face characteristic point position in this facial image, comprises eyebrow, canthus, nose, the corners of the mouth;
The human face characteristic point position of step (4), the facial image finally exported according to step (3) and correspondence thereof, choose two fixing human face characteristic points as fixed point, Rotation and Zoom process is carried out to facial image, export the facial image after alignment, in the facial image after alignment, the position of selected fixed characteristic points is by constant.
2. method according to claim 1, it is characterized in that: the zoom factor 0.7937 of the described multi-scale image pyramid model of step (1) had both contributed to the face that algorithm detects different scale size, was unlikely to again the computing time significantly increasing algorithm.
3. method according to claim 1, it is characterized in that: the non-face sorter of face described in step (2) is that input training obtains by extracting the feature of image, the feature of image to be merged mutually with convolutional neural networks feature by traditional characteristic and forms.
4. method according to claim 1, it is characterized in that: it be feature by extracting image is that input training obtains that human face characteristic point described in step (3) returns device, the feature of described image to be merged mutually with convolutional neural networks feature by traditional characteristic and forms.
5. method according to claim 1, is characterized in that: carry out Rotation and Zoom to facial image described in step (4), is to process based on the guarantee face left and right tail of the eye 2 is constant.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510983142.9A CN105512638B (en) | 2015-12-24 | 2015-12-24 | A kind of Face datection and alignment schemes based on fusion feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510983142.9A CN105512638B (en) | 2015-12-24 | 2015-12-24 | A kind of Face datection and alignment schemes based on fusion feature |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105512638A true CN105512638A (en) | 2016-04-20 |
| CN105512638B CN105512638B (en) | 2018-07-31 |
Family
ID=55720605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510983142.9A Active CN105512638B (en) | 2015-12-24 | 2015-12-24 | A kind of Face datection and alignment schemes based on fusion feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105512638B (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295579A (en) * | 2016-08-12 | 2017-01-04 | 北京小米移动软件有限公司 | Face alignment method and device |
| CN106446862A (en) * | 2016-10-11 | 2017-02-22 | 厦门美图之家科技有限公司 | Face detection method and system |
| CN106599883A (en) * | 2017-03-08 | 2017-04-26 | 王华锋 | Face recognition method capable of extracting multi-level image semantics based on CNN (convolutional neural network) |
| CN106650615A (en) * | 2016-11-07 | 2017-05-10 | 深圳云天励飞技术有限公司 | Image processing method and terminal |
| CN106909909A (en) * | 2017-03-08 | 2017-06-30 | 王华锋 | A kind of Face datection and alignment schemes based on shared convolution feature |
| CN107609459A (en) * | 2016-12-15 | 2018-01-19 | 平安科技（深圳）有限公司 | A kind of face identification method and device based on deep learning |
| CN107766774A (en) * | 2016-08-17 | 2018-03-06 | 鸿富锦精密电子(天津)有限公司 | Face identification system and method |
| CN107992833A (en) * | 2017-12-08 | 2018-05-04 | 北京小米移动软件有限公司 | Image-recognizing method, device and storage medium |
| CN108062542A (en) * | 2018-01-12 | 2018-05-22 | 杭州智诺科技股份有限公司 | The detection method for the face being blocked |
| CN108090468A (en) * | 2018-01-05 | 2018-05-29 | 百度在线网络技术（北京）有限公司 | For detecting the method and apparatus of face |
| CN108121950A (en) * | 2017-12-05 | 2018-06-05 | 长沙学院 | A kind of big posture face alignment method and system based on 3D models |
| CN108446672A (en) * | 2018-04-20 | 2018-08-24 | 武汉大学 | A kind of face alignment method based on the estimation of facial contours from thick to thin |
| CN108960029A (en) * | 2018-03-23 | 2018-12-07 | 北京交通大学 | A kind of pedestrian diverts one's attention behavioral value method |
| CN109902631A (en) * | 2019-03-01 | 2019-06-18 | 北京视甄智能科技有限公司 | A kind of fast face detecting method based on image pyramid |
| CN109902665A (en) * | 2019-03-28 | 2019-06-18 | 北京达佳互联信息技术有限公司 | Similar face retrieval method, apparatus and storage medium |
| WO2020056688A1 (en) * | 2018-09-20 | 2020-03-26 | 华为技术有限公司 | Method and apparatus for extracting image key point |
| CN111222436A (en) * | 2019-12-30 | 2020-06-02 | 广州市昭云互联网科技有限公司 | Image mixing identification method and system |
| US11200451B2 (en) | 2018-11-14 | 2021-12-14 | Canon Kabushiki Kaisha | Object recognition method and apparatus |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778448A (en) * | 2015-03-24 | 2015-07-15 | 孙建德 | Structure adaptive CNN (Convolutional Neural Network)-based face recognition method |
| CN105005774A (en) * | 2015-07-28 | 2015-10-28 | 中国科学院自动化研究所 | Face relative relation recognition method based on convolutional neural network and device thereof |
| US20150363634A1 (en) * | 2014-06-17 | 2015-12-17 | Beijing Kuangshi Technology Co.,Ltd. | Face Hallucination Using Convolutional Neural Networks |
-
2015
- 2015-12-24 CN CN201510983142.9A patent/CN105512638B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150363634A1 (en) * | 2014-06-17 | 2015-12-17 | Beijing Kuangshi Technology Co.,Ltd. | Face Hallucination Using Convolutional Neural Networks |
| CN104778448A (en) * | 2015-03-24 | 2015-07-15 | 孙建德 | Structure adaptive CNN (Convolutional Neural Network)-based face recognition method |
| CN105005774A (en) * | 2015-07-28 | 2015-10-28 | 中国科学院自动化研究所 | Face relative relation recognition method based on convolutional neural network and device thereof |
Non-Patent Citations (2)
| Title |
|---|
| KUSHSAIRY KADIR 等: "A comparative study between LBP and Haar-like features for Face Detection using OpenCV", 《 ENGINEERING TECHNOLOGY AND TECHNOPRENEUSHIP》 * |
| 高艳霞: "基于Haar-Lbp特征和粒子群神经网络的人耳识别", 《怀化学院学报》 * |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295579A (en) * | 2016-08-12 | 2017-01-04 | 北京小米移动软件有限公司 | Face alignment method and device |
| CN106295579B (en) * | 2016-08-12 | 2019-10-15 | 北京小米移动软件有限公司 | Face alignment method and device |
| CN107766774A (en) * | 2016-08-17 | 2018-03-06 | 鸿富锦精密电子(天津)有限公司 | Face identification system and method |
| CN106446862A (en) * | 2016-10-11 | 2017-02-22 | 厦门美图之家科技有限公司 | Face detection method and system |
| CN106650615A (en) * | 2016-11-07 | 2017-05-10 | 深圳云天励飞技术有限公司 | Image processing method and terminal |
| CN106650615B (en) * | 2016-11-07 | 2018-03-27 | 深圳云天励飞技术有限公司 | A kind of image processing method and terminal |
| CN107609459B (en) * | 2016-12-15 | 2018-09-11 | 平安科技（深圳）有限公司 | A kind of face identification method and device based on deep learning |
| CN107609459A (en) * | 2016-12-15 | 2018-01-19 | 平安科技（深圳）有限公司 | A kind of face identification method and device based on deep learning |
| CN106599883A (en) * | 2017-03-08 | 2017-04-26 | 王华锋 | Face recognition method capable of extracting multi-level image semantics based on CNN (convolutional neural network) |
| CN106909909A (en) * | 2017-03-08 | 2017-06-30 | 王华锋 | A kind of Face datection and alignment schemes based on shared convolution feature |
| CN106599883B (en) * | 2017-03-08 | 2020-03-17 | 王华锋 | CNN-based multilayer image semantic face recognition method |
| CN108121950B (en) * | 2017-12-05 | 2020-04-24 | 长沙学院 | Large-pose face alignment method and system based on 3D model |
| CN108121950A (en) * | 2017-12-05 | 2018-06-05 | 长沙学院 | A kind of big posture face alignment method and system based on 3D models |
| CN107992833A (en) * | 2017-12-08 | 2018-05-04 | 北京小米移动软件有限公司 | Image-recognizing method, device and storage medium |
| CN108090468B (en) * | 2018-01-05 | 2019-05-03 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detecting face |
| CN108090468A (en) * | 2018-01-05 | 2018-05-29 | 百度在线网络技术（北京）有限公司 | For detecting the method and apparatus of face |
| CN108062542B (en) * | 2018-01-12 | 2020-07-28 | 杭州智诺科技股份有限公司 | Method for detecting shielded human face |
| CN108062542A (en) * | 2018-01-12 | 2018-05-22 | 杭州智诺科技股份有限公司 | The detection method for the face being blocked |
| CN108960029A (en) * | 2018-03-23 | 2018-12-07 | 北京交通大学 | A kind of pedestrian diverts one's attention behavioral value method |
| CN108446672A (en) * | 2018-04-20 | 2018-08-24 | 武汉大学 | A kind of face alignment method based on the estimation of facial contours from thick to thin |
| CN108446672B (en) * | 2018-04-20 | 2021-12-17 | 武汉大学 | Face alignment method based on shape estimation of coarse face to fine face |
| WO2020056688A1 (en) * | 2018-09-20 | 2020-03-26 | 华为技术有限公司 | Method and apparatus for extracting image key point |
| US11200451B2 (en) | 2018-11-14 | 2021-12-14 | Canon Kabushiki Kaisha | Object recognition method and apparatus |
| CN109902631A (en) * | 2019-03-01 | 2019-06-18 | 北京视甄智能科技有限公司 | A kind of fast face detecting method based on image pyramid |
| CN109902631B (en) * | 2019-03-01 | 2021-02-26 | 北京视甄智能科技有限公司 | Rapid face detection method based on image pyramid |
| CN109902665A (en) * | 2019-03-28 | 2019-06-18 | 北京达佳互联信息技术有限公司 | Similar face retrieval method, apparatus and storage medium |
| US11294953B2 (en) | 2019-03-28 | 2022-04-05 | Beijing Dajia Internet Information Technology Co., Ltd. | Similar face retrieval method, device and storage medium |
| CN111222436A (en) * | 2019-12-30 | 2020-06-02 | 广州市昭云互联网科技有限公司 | Image mixing identification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105512638B (en) | 2018-07-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105512638A (en) | Fused featured-based face detection and alignment method | |
| Zhou et al. | Face and gender recognition system based on convolutional neural networks | |
| Tang et al. | View-independent facial action unit detection | |
| CN103824052A (en) | Multilevel semantic feature-based face feature extraction method and recognition method | |
| Sharma et al. | Recognition of single handed sign language gestures using contour tracing descriptor | |
| Rai et al. | Gender classification techniques: A review | |
| CN109977958A (en) | A kind of offline handwritten form mathematical formulae identification reconstructing method | |
| CN111126240A (en) | Three-channel feature fusion face recognition method | |
| Shelke et al. | A novel multi-feature multi-classifier scheme for unconstrained handwritten devanagari character recognition | |
| Choudhary et al. | A new approach to detect and extract characters from off-line printed images and text | |
| Nasr et al. | Face recognition system using bag of features and multi-class SVM for robot applications | |
| Upadhyay et al. | A review on different facial feature extraction methods for face emotions recognition system | |
| Karahan et al. | Age and gender classification from facial features and object detection with machine learning | |
| Aly et al. | Arabic sign language recognition using spatio-temporal local binary patterns and support vector machine | |
| Jasim et al. | A real-time computer vision-based static and dynamic hand gesture recognition system | |
| Gohil et al. | Chain code and holistic features based OCR system for printed devanagari script using ANN and SVM | |
| Ismail et al. | Online Arabic handwritten character recognition based on a rule based approach | |
| CN104166840A (en) | Focusing realization method based on video conference system | |
| Karthik et al. | Segmentation and recognition of handwritten kannada text using relevance feedback and histogram of oriented gradients–a novel approach | |
| Rajithkumar et al. | Template matching method for recognition of stone inscripted Kannada characters of different time frames based on correlation analysis | |
| Gadge et al. | Recognition of Indian Sign Language characters using convolutional neural network | |
| Ajao et al. | Yoruba handwriting word recognition quality evaluation of preprocessing attributes using information theory approach | |
| CN110163136B (en) | Fingerprint and finger vein bimodal identification decision fusion method based on perceptron | |
| Sawant et al. | Word Transcription of MODI Script to Devanagari Using Deep Neural Network | |
| Duth et al. | Recognition of hand written and printed text of cursive writing utilizing optical character recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Wang Huafeng Inventor after: Pan Haixia Inventor after: Lv Weifeng Inventor after: Huang Jiang Inventor after: Cai Yehe Inventor after: Song Wenfeng Inventor after: Du Junyi Inventor before: Huang Jiang Inventor before: Wang Huafeng Inventor before: Cai Yehe Inventor before: Song Wenfeng Inventor before: Du Junyi Inventor before: Lv Weifeng |
|
| CB03 | Change of inventor or designer information | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20170627 Address after: 100191 Haidian District, Xueyuan Road, No. 37, Applicant after: Wang Huafeng Applicant after: Pan Haixia Applicant after: Lv Weifeng Applicant after: Huang Jiang Applicant after: Cai Yehe Applicant after: Song Wenfeng Applicant after: Du Junyi Address before: 100191 Haidian District, Xueyuan Road, No. 37, Applicant before: Huang Jiang Applicant before: Wang Huafeng Applicant before: Cai Yehe Applicant before: Song Wenfeng Applicant before: Du Junyi Applicant before: Lv Weifeng |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |