旋转人脸的模板匹配定位与旋转角度估算方法

Template Based Human Face Orientation Determination and Rotated Angle Estimation

本文实现了一种依靠非旋转的人脸模板对平面内旋转的人脸图象进行匹配定位与角度估算的方法。相对于通过由不同旋转角度的模板构成的集合中的成员分别作用于目标图象进行匹配定位的方法,本文的方法减轻了计算负担。而且本文将圆投影法与 Zernike 矩法相结合实现了旋转人脸图象的定位与旋转角度估算。其匹配的准确性比较高,而且具有良好的抗噪声特性。

In this paper, a method is proposed to determine the orientation of a rotated human face image as to be matched with a non-rotated face template. In contrast with conventional method in which templates with different rotated angles are tried one after another to be matched with the target image, a new method based on the combination of circle projection and Zernike moment is proposed. Using this method, computational complexity is reduced, match accuracy is improved, both rotated angle estimation and human face orientation can be determined simultaneously with stronger resistance against noise interference.