Abstract: In the dictionary learning algorithms based on Fisher criterion, the selection of initial dictionary, and the construction of the objective function, seriously influence the effect of dictionary learning. In order to reduce the influence of initial dictionary, improve the ability of expression and discrimination. A face recognition algorithm is presented based on Gabor features and adaptive weighted Fisher criteria. Firstly, Gabor filter is used to extraction face feature, and use the Gabor facial features as the face training sets. Secondly, by adding the attenuation function, and adaptive weighting the training set, improve the effect of the Fisher criterion dictionary learning algorithm. Finally, the coefficient errors of the test sample coding, are used to identify the category. The experimental results on AR and Extend Yale database show that this algorithm is not only a very good image information extract algorithm, but also can effectively improve the accuracy rate of face recognition.

Key words: dictionary learning, Gabor features, adaptive weighting, forgotten function, face recognition

摘要： 在基于Fisher准则的字典学习算法中，初始字典的选取和目标函数的构建，严重影响字典学习的效果。为了减少初始字典的影响，提高算法的表达和判别能力。提出了一种结合Gabor特征和自适应加权Fisher准则的人脸识别算法。该算法首先采用Gabor滤波器提取人脸特征，将提取到的Gabor人脸特征作为人脸训练集；通过添加遗忘函数和根据样本间的距离对训练样本自适应加权，改进Fisher准则字典学习算法；利用测试样本编码系数的误差进行识别。在人脸库上的实验表明，算法不仅能很好地提取图像的特征信息，而且可以有效地提高人脸识别率。

关键词: 字典学习, Gabor特征, 自适应加权, 遗忘函数, 人脸识别