Gabor特征集结合判别式字典学习的稀疏表示图像识别

稀疏编码中字典的选择无论对图像重建还是模式分类都有重要影响,为此提出Gabor特征集结合判别式字典学习的稀疏表示图像识别算法.考虑到Gabor局部特征对光照、表情和姿态等变化的鲁棒性,首先提取图像对应不同方向、不同尺度的多个Gabor特征；然后将降维的增广Gabor特征矩阵作为初始特征字典,通过对该字典的学习得到字典原子对应类别标签的新结构化字典,新字典中特定类的子字典对相关的类具有好的表示能力,同时应用Fisher判别约束编码系数,使它们具有小的类内散度和大的类间散度；最后同时用具有判别性的重构误差和编码系数来进行模式分类.基于3个数据库的实验结果表明本文方法具有可行性和有效性.

Sparse representation for image recognition based on Gabor feature set and discriminative dictionary learning

Choosing the right dictionary used for sparse coding has an important effect on image reconstruction and pattern classification. Therefore, a new sparse representation algorithm based on Gabor Feature Set Discriminative Dictionary Learning is proposed for image recognition. Considering that Gabor feature is robust to variations of illumination, expression, and pose, the proposed method first extracts the image Gabor features with multi-scale and multi-orientation.Then it uses the augmented Gabor local feature matrix whose dimension has been reduced to construct the initial feature dictionary. This reduction is based on the Fisher discrimination criterion. A structural dictionary, whose atoms correspond to the class labels, is learned so that each sub-dictionary of the learned new dictionary is a good representation of the samples from the corresponding class. Meanwhile, the Fisher discrimination criterion is imposed on the coding coefficients so that they have small within-class scatter but big between-class scatter. Consequently, a new classification scheme associated with the proposed method is then presented by using, the discriminative information and sparse coding coefficients. Experiments on three types of databases show that the proposed method is valid and efficient.