欢迎访问《图学学报》
分享到：
摘 要：基于稀疏表示的人脸识别算法(SRC)识别率相当高，但是当使用l1范数求最优的稀疏表示时，大大增加了算法的计算复杂度，矩阵随着维度的增加，计算时间呈几何级别上升，该文提出利用拉格朗日算法求解矩阵的逆的推导思路，用一种简化的伪逆求解方法来代替l1范数的计算，可将运算量较高的矩阵求逆运算转变为轻量级向量矩阵运算，基于AR人脸库的实验证明，维度高的时候识别率高达97%，同时，计算复杂度和开销比SRC算法大幅度降低95%。
关 键 词：稀疏编码；分类方法；人脸识别；小波变换；快速算法
Abstract: As a recently proposed technique, sparse representation based classification (SRC) has been widely used for face recognition (FR). Sparse representation based SRC algorithm has a high recognition rate. While l1-minimization (l1-min) has recently been studied extensively in optimization, the high computational cost associated with the traditional algorithms has largely hindered their application to high-dimensional, large-scale problems. This paper devotes to analyze the working mechanism of SRC and discusses accelerated l1-min techniques using augmented Lagrangian methods, consequently, we propose a very simple yet much more efficient face classification scheme. The performance of the new algorithms is demonstrated in a robust face recognition of AR database. The experimental results verify that these methods can greatly improve the face recognition speed rate (97% decrease), and maintain a high recognition rate (95%). These methods are of practical values.#br# Key words: sparse representation; classification method; face recognition algorithm; gabor wavelet; fast algorithm
