| algorithm is more classical, it is a global feature-based extraction algorithm. Correspondingly, based on localized feature extraction algorithm, sparse coding algorithm that is based on local feature extraction algorithm, it has reduced redundant data, enhanced data robustness advantages. Pattern classifier can be divided into non-linear classifier with the linear classifier. SVM is a linear classifier, it is a small sample of data is the best classification results, are widely used in face recognition.Although the sparse coding algorithm has many advantages, but at this stage iterative sparse coding algorithm for a long time, less efficient, this paper presents an efficient sparse coding algorithm norm 0, the break point in the model is solved after continuous development, will greatly enhance the algorithm for computing efficiency. Due to the sparse coding algorithm applied to the face recognition, pattern classification stage is essential, so this stage also on the pattern classification support vector machines for simple research, the ultimate punishment for the model to improve the selection of parameters. End-use efficient sparse coding algorithm 0-norm combined with the improved support vector machine composed of a new efficient approach for face recognition.In order to verify the proposed 0-norm sparse coding efficiency of the algorithm, compared with NMFs algorithm, experiments on the ORL face database. The final experimental data show that the proposed 0-norm sparse coding algorithm is better than NMFs algorithm convergence rate, greatly reducing the overall iteration time. Penalty parameter of support vector machines have also been the choice of experiments and eventually got a penalty parameter of the method of choice, that attitude does not look rich face database that we can be relatively random selection correction coefficient matrix, a rich person profile of expression face database we choose g corresponding to a smaller correction coefficient matrix. |
PDF Full Text Request