Face correction and recognition of multi-pose based on Gaussian process regression

To solve the problem of the decline in success rate of face recognition with the change of facial attitude, We analyze the relationship of contour between frontal face and side face based on Gaussian process regression and propose a method to process side face with horizontal angle from −45° to +45°. We experiment in Multi-Pie and FERET database and the result shows the method in this paper significantly improves the success rate of side face recognition.