Study On Target Recognition Technique Based On Support Vector Machine

Target recognition is a focus issue in the pattern recognition and image processing field. For its widely practical application in the society, the research has an important significance. Support vector machine (SVM), based-on the structural risk minimization principle in the statistical learning theory, solve the over fitting problem effectively while other traditional pattern recognition methods could not. So SVM become the most popular classifier in recognition field. This dissertation summarizes the theory associated with the technology of target recognition and SVM in the beginning. It proves the validity of using Hu moments invariance with displacement, scale, rotation, illumination invariability as eigenvector of recognition. Construct a SVM classifier based on SVMlight algorithm and rbf kernel, then train the classifier with Hu moment invariance for recognizing. Through the contrastive experiment, it proves that SVM classifier has a obvious predominance to the traditional ones especially in small samples and high dimension problems. In the end, a target recognition system which is used in human face recognition is established, the system realize target capture, target detection, feature extraction and target recognition respectively. Through plenty of experiments on standard human face database, the results show that the technology of recognition based on SVM and Hu moment invariance is better than BP Neural Networks.