针对非受控场景下人脸识别率低的问题, 提出一种非控场景下基于主成分稀疏表示与低秩分解的人脸识别算法。首先通过核心基础信息平台收集的数据自构建基础人脸库, 然后采集课堂照片并对采样照片通过主成分稀疏表示和低秩分解算法进行分割, 最后以基础人脸库为样本进行匹配识别, 并将未进行低秩分解的情况与低秩分解后的结果进行比较。实验结果表明, 在非受控场景下通过主成分稀疏表示叠加低秩分解的识别效果对光照变化影响的鲁棒性较强, 对遮挡情况受到的影响相对明显, 总体识别正确率最高达到92.4%, 达到较好非控场景下人脸识别效果。该算法对开放型非受控场景下的人脸识别及由此衍生出的表情识别、行为识别等研究都有积极助益。

In order to solve the problem of low face recognition rate in uncontrolled scenes, a face recognition algorithm based on sparse principal component representation and low rank decomposition in uncontrolled scenes is proposed. Firstly, the basic face database is constructed by the data collected by the core basic information platform, and then the classroom photos are collected and the sampled photos are segmented by principal component sparse representation and low rank decomposition algorithm. Finally, the basic face database is used as a sample for matching recognition, and the results without low rank decomposition are compared with those after low rank decomposition. The experimental results show that the recognition effect of superimposed low rank decomposition by sparse representation of principal components in uncontrolled scene is robust to the change of light, and the influence on occlusion is relatively obvious. The highest recognition accuracy is 92.4%, which achieves a better face recognition effect in uncontrolled scenes. The algorithm is helpful to the research of face recognition, expression recognition and behavior recognition in open uncontrolled scenes.