With the rapid development of artificial intelligence technology, Face recognition becomes a hot research area in the field of computer vision because of its high academic research value and broad application prospects. Compared with other biometric technologies, Face recognition has the advantages of user-friendly process, easy operation and multi-people simultaneous recognition. Face recognition has been mainly used in attendance and security fields nowadays. Face recognition based sparse representation is a hot research topic in recent years and has the advantages of simple framework, insensitive feature selection and strong robustness, which cause the attention of domestic and foreign scholars. Although there are a large number of research achievement, how to improve the recognition accuracy in difficult condition is a problem needed to be solved. This dissertation focused the above problem and covered the following works:1. Propose a illumination-compensate dictionary learning method. To get a better recognition accuracy under complicated illumination, the original sparse representation algorithm need the training samples to contain sufficient illumination variation. To overcome the dependence of sparse representation on the sufficiency of training samples, We proposed a illumination-compensate dictionary learning method. This method learns a dictionary which get a best represent ability for illumination variation and we both use the illumination-compensate dictionary and training samples to linearly reconstruct the test sample in classification stage. Experimental results show that the proposed method achieves good classification accuracy with a few training samples.2. Propose a discriminative dictionary learning method. Whether the dictionary has a good discriminative ability or not is effect on the recognition accuracy of sparse representation algorithm. To enhance the recognition accuracy, we propose a discriminative dictionary learning method. The proposed method use the label information of samples to construct a label consistent error item and a sparse coding classification error item and introduce these error items into the target function to build a unified model for training dictionary, linear transformation matrix and classifier. The model can be solved by a iterative convergence method. Experiment results show that this method has a good classification performance.3. Propose a face recognition algorithm based on local region sparse representation. In order to deal with local occlusion, we proposed a face recognition algorithm based on local sparse representation. The algorithm divides a face image into several local regions and learning dictionary and classifier for each local region independently. Firstly, we calculate response value of each local region on test sample by corresponding dictionary and classifier. Secondly, we build occlusion masks according to the response values and use the mask to identify occlusion regions. Finally, we get the classification result by integrating the response values of all non-occlusion regions. |