Research On Method For Extraction Of Stable Visual Feature In Targe Recognition

The ultimate goal for the computer vision is to realize the function as biological visual recognition,as well be one of focus among researchs at home and abroad in recent years.How to effectively obtain information about the target is the key technology for target recognition.Among the many features of target,accurately selecting more intuitive,stable features of target can greatly increase the rate and accuracy for recognition.This paper made deep study in the theory of biological vision.Modeled on the biological visual choice theory,the paper proposed feature selecting principle which is based on visual stability.The paper provided a theoretical guidance for feature selection.The paper improved present methods for feature extraction and proves the feasibility of the principle which is put forward by the paper,while providing a new direction for visual research.Firstly,the paper delimited the stable visual feature that based on biological theory.According to the feature which is intuitive and stable,the paper exetracted some typical features.It improved multi-channel Gabor filtering and effectively extracted texture feature.Secondly proposed a multi-operator integration algorithm by combining Multi-resolution Analysis and wavelet entropy.This method selected main energy of the edges obtained by each operator, effectively obtain the smooth edge which is with no noise,continuous and complete from target.Next penetrated into the moment theory and summed up that the target image can supply stable moment feature which is the basis for identification.The paper also proved it by experiment.Finally,the paper introduced neural networks and the fuzzy theory.It proposed fuzzy BP neural neteork method to recognize images which are polluted by noise.The experiment proved that the method can realise accurate rate for recognition and with satisfactory.In summary,this paper definited the stable visual feature on the basis of biological, provided feature theoretical guidance for recognition,and proposed some new methods which are effevtive for recognition.All experimental results showed that the methods which are based on stable visual feature by the paper are effective for target recognition while has accurate results and with satisfactory.