Algorithm Research Based On Dictionary Learning And Its Application In Face Recognition

In the field of face recognition,the recognition accuracy of the face recognition algorithm is closely related to the number of training samples provided when training the algorithm.When the number of research samples provided is small,many traditional face recognition algorithms cannot maintain the recognition rate within the ideal level.Therefore,how to reduce the dependence of the algorithm on the number and quality of training data has always been the field of face recognition and even artificial intelligence One of the research hotspots.Based on the dictionary learning algorithm,this paper focuses on how to improve the accuracy rate of face classification by the dictionary learning algorithm when the training data is small and the data difference is large,so as to improve the robustness of the algorithm while maintaining high operating efficiency.In this paper,the improvement of dictionary learning algorithm mainly focuses on the following aspects.1.Most of the traditional dictionary learning methods focus on improving the feature extraction of different types of face data in dictionaries,ignoring the importance of extracting the common features of different types of face data.In order to ensure that the algorithm can extract the category difference and the difference commonness,a hybrid dictionary learning algorithm is proposed.The hybrid learning algorithm includes two parts: learning class-specific dictionary and learning intra-class feature dictionary.On the basis of identifying sub dictionaries,the class feature dictionary preserves the similarity of sparse coefficients as much as possible through Laplacian constraints,so that the sparse coefficients work better on the classifier.2.Aiming at the situation where the data samples are quite different,a fusion dictionary learning algorithm model is proposed.When the training data are mostly positive pictures of the face under normal light,and the test data are mostly the face occlusion pictures and posture changes under abnormal light,it is necessary to extract the interference of the data.In this paper,the fusion dictionary learning algorithm is proposed.Firstly,in order to improve the operation efficiency,the LBP feature pyramid is used to preprocess the data,and then the class-specific dictionary,the common dictionary and the perturbation dictionary are learned,so that when the fusion dictionary learning algorithm is classified,the data disturbance can be considered,and the classification can be better based on the class difference and the difference commonness.3.In order to improve the robustness of the algorithm and make better use of the influence of data disturbance on classification results,a sparse comprehensive dictionary learning algorithm is proposed.The algorithm is divided into three steps: hybrid characteristic dictionary learning,extended interference dictionary learning and low-rank dictionary learning.The sparse coefficient is constrained by Fisher criterion,and the sparse comprehensive dictionary and sparse coefficient are classified.The algorithm extracts the sparse components in the face image by low rank matrix decomposition.This paper designs a variety of simulation experiments in AR,CMU-PIE,YALE,YALEB,LFW and other face databases,and compares it with a variety of dictionary learningalgorithms.The experiment proves that the dictionary model proposed in this paper has a higher level of face data Recognition accuracy and algorithm robustness.