Research On Image Recognition Technology Based On Dictionary Learning

Image recognition is an important area of research in the field of computer vision and pattern recognition. With the rapid development of computer technology and information technology, image recognition technology has received many applications in medical,information security, industrial, astronomy and meteorology. Sparse representation has good robustness, strong anti-jamming capability and generalization ability, has been successfully applied to image recognition and become a hot research. On the basis of in-depth study of the sparse representation and dictionary learning theory, and apply them on image recognition in this paper.The thesis describes the research background, research significance and study status at home and abroad of the image recognition. And some classical algorithms of image recognition is introduced and analyzed in detail. Summarize on the theory of sparse representation and dictionary learning, and describes the K-SVD algorithm,discrimination K-SVD algorithm in detail and describes how they apply the dictionary learning algorithm to image recognition.Considering that Gabor feature is robust to variations of illumination, occlusion and expression, combining the theory of image recognition and discriminative dictionary learning algorithm, a face recognition algorithm was proposed based on Gabor feature and support vector guided dictionary learning(GSVGDL). The proposed algorithm extracts image Gabor features instead of the whole face image as input of a dictionary to learn. The algorithm is an improved discriminative dictionary learning algorithm. The dictionary learning model combined the reconstruction error with the discrimination term and the regularization term.And formulate the discrimination term as the weighted summation of the squared distances between all pairs of coding vectors. Then a discriminative dictionary and linear classifier was learned simultaneously by the dictionary learning, which the learned dictionary atoms were corresponded to the class labels. The dictionary learning algorithm improves the discriminative of encoding vector. Projection encoding of the test sample using the discrimination of the encoding vector, classification and recognition of the query image.Experimental results verify the effectiveness and the feasible of the algorithm on the face databases and handwritten digital database.