Face Recognition Based On Dictionary Optimization

In recent years,sparse representation theory has been widely used in the field of face recognition.And the key of face recognition based on sparse representation is to construct discriminative dictionaries.Therefore,the main research of this paper is to optimize the dictionary and improve the face recognitionBased on the theory of sparse representation,three kinds of dictionary optimization methods are proposed in this paper.?1?In view of the limited number of training samples and the interference,this paper proposes a face recognition algorithm based on united dictionary.The algorithm obtains a discriminative low-rank dictionary and a sparse error dictionary during the training phase,and then combines the two dictionaries as a dictionary for testing.The classification is classified according to the sparse coefficients corresponding to the discriminative low-rank class dictionaries.?2?Different from previous methods,this paper proposes a face recognition algorithm based on discriminative double dictionary learning,which can directly improve the dictionary's discriminability by introducing different kinds of expression errors to directly restrict the sparsity coefficients.Secondly,the analysis dictionary is trained in the training phase.In the test phase,Using an analytical dictionary to find the sparse coefficient is much faster than the traditional ₀l or ₁l norm method,?3?At present,the algorithms based on sparse representation are mostly separated from the data reduction and dictionary training,which may not be able to obtain the optimal discrimination information of the data.Therefore,this paper proposes a face recognition algorithm based on simultaneous dimensionality reduction and double dictionaries learning.Secondly,according to the discriminative sparse coding in LCKSVD algorithm,this paper proposes a method of using cosine similarity to optimize.In addition,the algorithm also trains analysis dictionary during the training phase,which will save time in the testing phase.Experimental results in AR and Extended Yale B database showed that the three dictionary optimization methods proposed are effective and robust,which are suitable for practical application.