Research On Algorithm Of Multi-Pose Face Correction And Recognition Based On ASM

With the development of smart phones,Internet finance,and the application of intelligent security robots,face recognition is becoming more and more widely used in identity authentication and crime monitoring.Due to the limitation of the photographing conditions,in many cases,the camera can not acquire the standard frontal face images.But the performance of the current face recognition algorithm in multi-pose is not ideal.the existing face recognition algorithm in multi-pose has the problems of high computational complexity and low recognition rate.To solve the problem of the low recognition rate of current face recognition algorithm in multi-pose situation,in this paper we proposed an algorithm to correct the multi-pose face image to the standard frontal face image and then do recognition to solve the recognition problem of images with horizontal deflection angle in [-45°,+45°].In this paper,three face classifiers are trained based on the AdaBoost algorithm,which are responsible for the detection of the left face,the right face,and the frontal face,then we connect these classifiers to detect the faces in the whole angle ranges.By dividing the face pose into five subintervals,and training the ASM in each subinterval to fit the faces with the similar pose in the same interval to get the shape contour of face.And combine it with feature points location to determin the pose subinterval the image belong to and choose the best matching ASM.Then,we study the mapping rule between the side face contour and the front al face contour using the Gaussian process regression.And we predicted the frontal face contour from the side face contour based on the mapping function.Since the contour of the frontal face are obtained,we overlaid the texture of the side face image into the predicted face contour by segment affine method and got the corrected face image.In order to test the effect of face correction,we recognize the corrected multi-pose face image from face databases based on LGBPHS.By comparing the recognition rate of face images before and after pose correction,The results showed that the recognition rate is obviously improved after pose correction.By comparing the data with the other researchers,the result showed the recognition rate is improved,which proves the effectiveness of our algorithm of multi-pose face correction.