Head Posture Detection Based On Geometric Feature Analysis For Identification Photos

With the rapid development of Internet,more and more people prefer to make ID photos on their own hands.Generally speaking,there are several steps to make ID photos: capture,clip,color optimization and background replacement.If user makes a ID photo which doesn't meet the standards,such as head skew or ears are covered,the photo will be unusable.Therefore,Posture detection should be applied to the original photo in advance,if it's unusable,inform user to capture again.The posture detection method in this paper contains two points: head skew detection and ear covered detection.The posture detection is based on geometrical feature analysis.For the head skew issue,firstly,locate the face organs,including eyes?nose and mouth;then construct some features according to the prior knowledge of face's topological relationship;finally,extract the features from many samples to train a decision tree,so the head skew issue can be detected.For the ear covered issue,a parabolic curve based on the shape and geometry features of face contour,is used to fit the outer contour of skin near ear,whether ear is covered can be judged by the error of fitting;furthermore,this paper proposed a gradient and parabolic guidance based dynamic programming algorithm,to extract the face contours near ear in ID photos,on the premise of skin is detected,use the contours as benchmark,eliminate the useless part to get ear regions.At last,accurately determine whether the ear is covered and measure the degree of cover,by the quantitative analysis of ear region area and some other parameters.The main innovation points of this paper are:(1)proposed a gradient and parabolic guidance based dynamic programming algorithm,to extract the face contours near ear in ID photos.Firstly,the algorithm get a rectangular region contains ear,with the help of the position relations between eyes and ears;then,calculate gradient of the rectangular region;and design the energy function of contour,according to gradient and parabola guidance;finally,use dynamic programming method to solve the energy function,minimize the energy function,get the contour of face.(2)use the decision tree method to get the rules can judge whether head is skew,apply the existing theory to practical application,make the result more accurate.