The Research Of Sparse Coding And Its Applications In Image Classify

Biological exeperiments show that response properties of visual cortex have been considered to be sparse neuro-representation .Sparse coding has its origin in the study of visual neural network; it is a neural network method for finding a representation of multidimensional data in which each of the components of the representation is only rarely significantly active. Sparse code theory establishes a scientific quantitative link between the information processing mechanisms of visual neurons and the statistics of input visual stimuli, and provides an efficient tool to understand the neural information processing mechanisms. It has been applied in blind source separation, speech signal separation, image feature extraction; natural image denoising and pattern recognition. It has achieved many fruits and has important practical value.The sparse coding of natural image is an artificial neural network method, which can model the receptive fields of simple cells in the mammaliam primary visual cortex in brain. The encoding realization for this method only depends on the statistical properties of natural perceptive information, regardless of the inherent properties of input. It is a self-adaptive signal statistical method. This thesis presents the theories and algorithms of Sparse Coding and explores its applications to the natural image feature extraction. The main work of the author focuses on the following aspects.Firstly, this thesis is mainly about the research history and current development of the Sparse Coding Theory. It introduces the basic knowledge of statistic and information relating to the Sparse Coding Theory, idea and mathematical model of traditional supervised sparse coding,and also describing sparse coding model of the natural image in mathematics.Secondly, it puts forward the idea of setting up the coding structure, training a special dictionary for every class objects, and states its advance in training time, robustness and the efficiency comparing with the algorithm of tradition supervised sparse coding. At the same time it shows the experimental result and analysis. On the other hand, it puts forward the analysis methods based on the distribution situation of sparse feature sample in the feature space. Thesis introduces the Mean Shift algorithm, for the right problem of object identification puts forward two kinds of pretreatments: normalization, l0_regularization. Analyzing and comparing the different bias of the two methods theoretically. On the base of the situation in the sample distribution of feather space, it proposes object region extraction algorithm.Thirdly, achieve natural image object discernment system applying the Sparse Coding Theory. The system depends on the data in VOC2008 image warehouse and applies multi-dictionary structure mentioned in the paper.In conclusion, this paper deeply researches the algorithm of sparse coding and questions of object identification, and also discusses further direction of research in this area.