Survey On Object Recognition Based On Feature Extraction Using Independent Comonent Analysis

Object recognition is an important research problem which is is significant in both theoretical meanings and prcatical values. It has been put into use widely in the scopes of surveillance systems, medical diagnosis, military object tracking and so on. However, Object recognition is very difficult to deal with as a result of the diversity and complexity of targets themselves. Besides, the issue my be greatly influenced by the real-time enviromental factors such as weather status,illumination changes, shift of the camera, disturbance of noises. The difference of images of targets not only behaves the overall features indicated by the whole contours of targets,the difference of the local characteristics reflected by the edge details of them is more fundmental. Therefore, based on the high level statistics of image data of targets, the method of independent component analysis is proposed in the thesis. It is capable of extracting both the local and overall characteristics of the target images, which are used to establish eigen-spaces for the purpose of recognition. The method of ICA improves the deficiency of some traditional methods extracting overall features based on 2-order statistics such as PCA and SVD. In this paper, the basic principles of independent component analysis are discussed detailedly firstly.Secondly,two kinds of effectual algorithms in common use are put forward, Infomax ICA and FastICA included. At last, this paper deals with the applications of the technique of feature extraction using independent component analysis in the field of object recogntion particularly, and some simulant experiments are performed at same time. In the experiment of object recognition,in the light of the disorder and redundancy of the enormous features extracted by the method of ICA, some certain improvement is perfomed. To identify the type of targets acurrately just with few features is realized by the improved means. In addtion, the formula of distance calculation is changed correspondly, increasing the rate of object recogntion furtherly. The results of the experiments above proves that the method discussed in this paper has an strong stablity for object recognition toward such changes as pose alteration, disturbance of noise, illumination changes and so on. Compared with the traditional techniques of feature extraction, the method of feature extraction based on ICA exhibit greater andvantages for the recognition of targets with different local charactersitics and the same overall features.