基于多尺度梯度相角的人脸图像对称性检测与定位

Face Image Symmetry Detection and Location Based on Multi-scale Gradient Angular

在利用反对称双正交小波变换系数获得人脸图像多尺度梯度矢量的基础上 ,提出了一种基于多尺度梯度相角的人脸图像对称性检测与定位算法 ,该方法对光照变化具有很强的鲁棒性 .另外 ,本方法可利用小波变换的多分辨率特性 ,实现由粗至精的分级检测 ,从而提高定位速度 .同时 ,由于反对称双正交小波还能用于图像压缩 ,因此该方法为压缩数据域中的人脸定位提供了一条可能的途径 .与类似方法的对比实验表明 ,本文方法在定位速度。

A novel approach to face image symmetry detection and location is given using multi-scale gradient angle obtained from anti-symmetrical biorthogonal wavelets (ASBW) decomposition coefficients. The method is robustness to different illuminant conditions. Because we compute the gradient angle using ASBW, a hierarchical symmetry detection procedure can be easily implemented. In addition, our method provides a way to locate face in compressed data domain if the image is coded by an ASBW. Compared to the similar method, our method can yield results with higher location accuracy and decreasing computational cost. Comparative experimental results are given using MIT database.