基于类独立核稀疏表示的鲁棒人脸识别

Robust Face Recognition Based on Class Dependent Kernel Sparse Representation

针对人脸识别中光照变化、噪声干扰和遮挡等导致识别率下降的问题,提出类独立核稀疏表示的分类算法。利用冗余字典由多个子字典构成的特点,引入核技术用于提高人脸识别率。应用各类子字典和误差矩阵建立类独立核稀疏表示模型,借鉴正交匹配追踪算法思想提出类独立核正交匹配追踪算法,用于求解该模型得到各类的稀疏表示系数。将该系数结合各类子字典计算类相关重构误差,实现测试样本的分类识别。实验结果表明,相比同类算法,该算法具有较高的识别率,鲁棒性较好,能够有效抑制噪声、光照以及遮挡等干扰带来的负面影响。

For the problem of face recognition with noise,illumination,and occlusion,Class Dependent Kernel Sparse Representation based Classification（CDKSRC）is proposed.The basic idea is that redundant dictionary is composed of many sub-dictionaries and kernel technology is used to improve face recognition rate.CDKSRC model is constructed by each class sub-dictionary and error matrix.Using the basic idea of Orthogonal Matching Pursuit（OMP）,Class dependent Kernel Regularized Orthogonal Matching Pursuit（KROMP）technology is proposed to solve this model to obtain sparse representation coefficients.The reconstruction error associated with the each class can be calculated to achieve classification of the test sample by the sparse coefficients and each class sub-dictionary.Compared with state-of-the-art methods,the proposed algorithm gets a higher recognition rate,while it has good robust to noise,illumination,and occlusion,etc.Experimental results validate the effectiveness of the proposed algorithm.