融合图像梯度方向的客户相关算法分析

Analysis of Client-related Algorithm Fusing Image Gradient Orientations

针对人脸图像受光照变化影响导致大部分传统的依赖图像表征信息的子空间学习算法鲁棒性差这一问题,在图像梯度方向和客户相关技术的研究基础上,提出一种融合图像梯度方向的客户相关算法(CS-IGO-LDA).采用图像梯度方向来代替像素强度表示原始样本,并用客户相关方法提取每一个不同个体的样本特征向量以更好地描述不同类别之间的差异.提出的CS-IGO-LDA方法充分利用了图像梯度和客户相关方法在人脸识别中的优势.在XM2VTS人脸库上的实验证明了新算法在人脸验证方面的有效性.

Most of the traditional appearance-based subspace learning algorithms have poor robustness because human face images are affected by light changes.Based on studies of image gradient orientations and client-related technology,a client-related algorithm fusing image gradient orientations(CS-IGO-LDA)was proposed.For a better description of the discrepancies between different classes,the original sample was represented by image gradient orientations instead of pixel intensity,and client-related algorithm was introduced to extract sample feature vector of each individual.The CS-IGO-LDA makes full use of the merits of IGO and client-related technique.The results obtained from XM2VTS face database show the effectiveness of the new method on face verification.