协同表示算法对人脸图像具有快速分类的特点, 但在单样本或欠样本情况下, 对变化复杂的人脸识别率还不够理想, 无法满足工程要求。针对该问题, 提出一种基于欠样本混合内变基字典的 扩展协同表示算法。首先借助在同一环境下采集到的不同人脸的变化过程, 提取人脸的变化共同特征并生成内变基, 再融合两种及两种以上不同人脸变化的共同特征生成混合内变基, 提高内变基的 通用性, 建立训练样本与测试样本之间变化的稀疏字典。训练样本在字典帮助下近似构建出测试样本的特征人脸, 达到扩展训练样本集的目的, 再对人脸协同分类。利用AR库、ORL库、Yale库和 Yale B库进行识别实验。结果表明, 本文算法能有效提高协同表示算法的识别率, 在欠样本情况下识别率提高7.33%～3317%, 在单样本情况下识别率提高6.78%~24.47%。

The cooperative representation algorithm has the characteristics of rapid classification of face images, but in the case of single or undersample, the complex change of face recognition rate is not ideal enough to meet the engineering requirements. For this problem, a cooperative representation of face recognition algorithm with a mixed internal variable-base sparse dictionary is proposed. Firstly, with the help of the change process of different faces collected in the same environment, the common features of the change of the face are extracted and the invariant basis is generated, the generality of the invariant basis generated by the common features of two or more different changes of the face is improved, and the sparse dictionary of the change between the training sample and the test sample is established. With the help of the dictionary, the training samples can construct the feature faces of the test samples approximately, so as to expand the training sample set, the characteristic face of the test sample is constructed. Using the AR, ORL, Yale and Yale B library for identification experiments, the results show that this algorithm can effectively improve the recognition rate of the cooperative representation algorithm, and increase the recognition rate by 7.33% to 33.17% in the case of undersamples, and 6.78% to 24.47% in the case of single samples.