基于稀疏表示的人脸图像高光检测和消除

Specular Detection and Removal on Human Face by Sparse Representation

在图像处理中,人脸图像的分析与处理是一个重要问题,对于进一步的视觉任务如人脸识别等具有重要的意义。然而,由于人脸皮肤材质由较薄的表皮层和较厚的真皮层组成,而且表皮层包含油脂。因此,人脸皮肤在较强光照条件下可能出现高光,造成传统的人脸图像方法在处理带有高光的人脸图像时失效。利用高光分布的稀疏性,将高光检测问题转化为带有稀疏性约束的非负矩阵分解。进一步利用L1范数的近似对输入图像进行预处理,加速非负矩阵分解的过程。并且,将检测出的高光区域视为信息丢失,利用基于稀疏表示的图像修复方法恢复出高光区域的漫反射信息,达到消除高光分量的目的。

In image processing,the analysis and processing of face images is an important issue,which is of great significance for further visual tasks such as face recognition.However,since the skin material of the face is composed of a thin skin layer and a thicker dermis layer,and the skin layer contains grease.Therefore,the face skin may appear highlight under strong lighting conditions,causing the conventional face im.age method to fail when processing the face image with highlights.Using the sparsity of the high-light distribution,the high-light detection problem is transformed into a non-negative matrix decomposition with sparsity constraints.The input image is preprocessed by using the approximation of the L_1 norm to accelerate the process of non-negative matrix factorization.Moreover,the detected highlight region is re.garded as information loss,and the image reflection method based on the sparse representation is used to recover the diffuse reflection infor.mation of the highlight region,thereby achieving the purpose of eliminating the highlight component.