摘要
为了增强人脸识别对光照变化的鲁棒性,提出了一种融合多方法的人脸图像光照预处理算法。该算法首先根据改进的自适应平滑算法(IAS)估计出原图像的亮度分量L,再用Retinex算法求得反射分量R,同时对原图像进行局部对比度增强(LCE)处理来增强图像细节;然后采用基于标准差(SD)的加权方法将多种方法有效融合起来;最后采用基于稀疏表示的分类(SRC)算法进行判别归类。在Yale B人脸库上的实验表明,构造的算法识别率高于使用单一预处理算法,而且在训练样本单一、光照环境较差情况下也能取得很好的识别效果,对光照变化有较好的鲁棒性。
In order to enhance the robustness of face recognition to illumination change, an illumination preprocessing algorithm of face image with fusing several algorithms is proposed. Firstly, the luminance component L is estimated from the original image according to the improved adaptive smoothing （IAS） algorithm, then reflection components R is obtained using Retinex algorithm. At the same time, the local contrast enhancement （LCE） algorithm is used to enhance image details. And the reweighted method based on the standard deviation （SD） is also adopted to calculate the weight and combine several algorithms effectively. Finally, sparse representation based classification（SRC） is used to classify. The experiment results on the Yale B face databases show that the pro- posed algorithm has higher recognition rate than the single pretreatment algorithm, and in the single training sample and poor light- ing condition, this method can also achieve good recognition result, and has better robustness to illumination change.
出处
《电子技术应用》
北大核心
2015年第5期152-155,共4页
Application of Electronic Technique
基金
广西自然科学基金项目(2014GXNSFDA118035
2013GXNSFAA019331)
桂林电子科技大学研究生教育创新计划资助项目(GDYCSZ201462)
关键词
人脸识别
稀疏表示
自适应平滑
局部对比增强
标准差
face recognition
sparse representation
adaptive smoothing
local contrast enhancement
standard deviation
