Bio-inspired Invariant Feature Representation Of Images

The extraction and representation of invariant image features have been the emphasis and difficulty of pattern recognition and classification. It is well-known, however, that it is very easy for human beings to do this. Therefore, how to model the visual perception mechanisms for image features extraction and representation has both important theoretical and applied values. This thesis mainly focuses on bio-inspired invariant features representation method of images. The main work of this thesis is listed as follows:1. The brain mechanisms of visual information processing are summarized. The sparse code mechanisms in the brain are analyzed. Several visual system modeling methods are studied.2. Aiming at the problems of randomly extracting prototypes and huge computation existed in the SMF(Standard Model Features) model, a scheme of introducing fast dictionary learning algorithm into the SMF model is proposed. The proposed algorithm improves the accuracy of prototypes, and significantly accelerates the solutions to dictionary learning.3. A novel feature computation method combining SURF(Speeded Up Robust Features)algorithm and SMF model is presented here. The rotation- and scale- invariant features are achieved by the SURF algorithm. The proposed algorithm can effectively extra features, and solve complicated problems such as image rotation, scale and displacement changes.Experimental results on the benchmark data set demonstrate the effectiveness of the proposed method, and it achieves the best performance for image classification.