| Object recognition in the scene is a major research problem of imageunderstanding, it has important theoretical research value and application prospect.Becaues objects in the real world are variability, such as they have different scale,rotation, illumination, location, obstruction, etc. object recognition in complex andchaotic visual scene are both difficult for biology and machine vision system. So itbecame a hot research for current researchers that using biological intelligentinformation processing methods to guide object recognition. Through analyzingsome of the latest research results of current neuroscience and cognitive science,then based on the current existing classic biological vision system models, thisthesis carried out some related researches.The main work of this thesis includes:(1)Overviewed hierarchical visual perception mechanism in the primate visualcortex, given some of the current latest research of brain science, physiology,psychology, anatomy, cognitive science, etc. we used them as the theoretical basisof the established model.(2)Analyzed and realized some of the current existing biological visionmodels that based on primate visual cortex, and analyzed their own recognitionperformance as well as their possible applications, then gave their comparativeresults.we used them as the model basis of the established model.(3)Focus on the analysis of HMAX shape model, and draw on the idea ofmulti-feature combination of SEEMORE model, by adding color feature and texturefeature to the HMAX shape model, combine all features to guide object recognition,designed a feed-forward multi-feature fusion model for object recognition.(4)Verified object recognition performance of the feed-forward multi-featurefusion model that using SVM、NN、Adaboost three classifiers or SVM classifierwith different kernel function settings. Verified object recognition performance ofthe feed-forward multi-feature fusion model that on the Caltech5and GRAZ imagedataset. |
PDF Full Text Request