Research Of Reconstructing Facial Surface With Expression From Monocular Image

In the field of human-computer interaction and pattern recognition human face andfacial expressions related research have important theoretical value and a great prospect.These research have been subject to extensive attention. In the Past, whether facerecognition or facial expression recognition research are limit in the3D face. In furtherresearch, the lack of face information in2D image would subject to great constraints andlimitations. Such as it can not be effectively applied in facial expression recognition withdifferent postures. There are a lot of monocular images or monocular videos in the realworld. Therefore, in the field of facial expression analysis, single image with expressiveface3D information reconstruction technology can make effective use of existing imagesand video to provide better information for expression recognition and others face relatedtechnical sources. So they have been subject to more and more attention.Reconstruction shape from shading technique for human face use little CPU time andhas little constriant for the human face database. But this technology can not reconstruct3D human face with expression accurately. The use of the gradient-based constraints andcontrol point method would solve this problem, as follows:Gradient calculation of the face surfaces. The gradient constraint condition with athreshold value and the gradient difference based smoothness constraint conditions areused to improve the accuracy of the gradient estimate.Automatic control points positioning method. Based on motion estimation algorithmand threshold segmentation algorithm, a method is proposed to find the control points withno change in the process of expression evolution.Reconstruct3D surface based on the weighted least squares method and controlpoints. The problem with the weighted least squares method in the condition of greatdistance is pointed out. And an iterative weighted least squares method is proposed tosolve this problem. The theoretical analysis and experimental results show that theiterative weighted least squares method can maintain the gradient height constraint of thecontrol point and gradient constraints on the shape.