特征加权组稀疏判别投影分析算法

发布时间：2019-05-12 03:13

【摘要】：近来,稀疏表示分类算法已经在模式识别和特征提取领域获得了广泛的关注.受最近提出的稀疏表示判别投影算法启发,本文提出了一种新的特征加权组稀疏判别投影算法(Feature weighted group sparse classification steered discriminative projection,FWGSDP).首先,提出特征加权组稀疏分类算法(Feature weighted group sparsebased classification,FWGSC)进行稀疏系数编码,该算法采用带特征加权约束的保局性信息,能够鲁棒地重构给定的输入数据;其次,通过类内重构散度最小、类间重构散度最大为目标计算最优投影判别矩阵,使得输入数据具有最佳的模式分类效果;最后,提出迭代重约束稀疏编码方法并结合特征分解操作进行FWGSDP模型高效求解.在Ex Yale B,PIE和AR三个人脸数据库的实验验证了所提算法在普通数据和带噪数据中的分类效果都优于现存的算法.
[Abstract]:Recently, sparse representation classification algorithm has received extensive attention in the field of pattern recognition and feature extraction. Inspired by the recently proposed sparse representation discriminant projection algorithm, a new feature weighted group sparse discriminant projection algorithm (Feature weighted group sparse classification steered discriminative projection,FWGSDP) is proposed in this paper. Firstly, a feature weighted group sparse classification algorithm (Feature weighted group sparsebased classification,FWGSC) is proposed for sparse coefficient coding. The algorithm adopts the local information with feature weighted constraints and can robust reconstruct the given input data. Secondly, the optimal projection discriminant matrix is calculated by the minimum intra-class reconstruction divergence and the maximum inter-class reconstruction divergence, so that the input data has the best pattern classification effect. Finally, an iterative reconstrained sparse coding method is proposed and combined with feature decomposition operation to solve the FWGSDP model efficiently. The experiments of pie and AR face databases in Ex Yale B verify that the proposed algorithm is superior to the existing algorithms in both ordinary data and noisy data.
【作者单位】：浙江工业大学计算机科学与技术学院;衢州职业技术学院信息工程学院;
【基金】：国家自然科学基金(61502424;61379123) 浙江省自然科学基金(LY15E050007;LY15F030014;LQ14F030003)资助~~
【分类号】：TP311.13

【相似文献】