| Face identifiction is an important reasearch field in pattern recognition and image processing. Because of its non-infringement of personal privacy and non-contact and other unique advantage, face recognition has been widely applied in the field of security monitor and artificial intelligence. The focus of this thesis is the recognition algorithm of the facial feature extraction and classifier design for human face identifiction. By taking the recent development in machine learning, this dissertation has thoroughly studied the face recognition based on sparse representation and feature selection.This paper studies firstly two face recognition methods based on the compressive sensing theory, PCA and SRC. While based on the original classification of sparse characterization, the fractional matrix norm 2,( 0 1)pl <p < is proposed, which ultimately improves the recognition rate of face recognition.Mixed fractional matrix norm 2,( 0 1)pl <p ￡ is 2,( 0 1)pl <p ￡-based minimization. Its natural idea is combining the smoothness of Euclid norm2 l and the sparsity of norm( 0 1)pl <p ￡. the fractional matrix norm 2,( 0 1)pl <p < has exhibited good performance for feature selection in high-dimensional data processing. A variety of experimental results have showed that the fractional matrix norm 2,( 0 1)pl <p < has better joint sparsity and stronger reliability than traditional vector norm 1l. According to the structural characteristics of face data, a 2,( 0 1)pl <p ￡-based minimization is presented to select feature in this paper. For arbitrary parameters p ?( 0, 1], the consistency algorithm for solving this minimizing problem is designed, and this paper proved that the target function was strictly descending, which guarantees the convergence of the algorithm.. Combined with the nearest neighbor classification, a robust face recognition method is proposed. The extensive experiments on face data sets have showed that the 2,( 0 1)pl <p <-minimization model performs better than the state-of-the-art. |
PDF Full Text Request