Pattern Recognition Based On Discriminative Common Vectors

In the field of computer science, researchers hope that computers have human intelligence, and they carried out researches for this, pattern recognition was proposed in this situation. In daily life, human, often need to make judgments, and these were normal things to human, but these were very complex pattern recognition problems to computers. The general method that we work on complicated problems was begining with the basics, then gradually in-depth, and ultimately achieving the desired research results. As the basis of pattern recognition, the researchs of classification have great significance to pattern recognition. Support Vector Machines (SVM) as a very popular classification algorithm, It has become a very important research field of pattern recognition in recent years.In this paper, we have studied several domestic and foreign modified class of SVM, and aiming at the bad anti-noise performance of SVM, Common Vectors (CVs) has been introduced to SVM, and three new modified class of SVM were presented. Then proved them through the experiments. The main work of this paper was summarized as follows:The first section is exordium. In this section we introduce the research status and application fields of pattern recognition. And reviews the algorithm of SVM.In the second section introduce two modified class of SVM were mainly studied, they are Minimum Class Variance Support Vector Machines (MCVSVMs), and Total Margin v Support Vector Machine (TM-v-SVM). Studied on the theoretical system of common vectors, and try to bring CVs into SVM.In the third section, Total Margin v Minimum Class Variance Vector Machines (TM-v-MCVSVMs) was presented. TM-v-MCVSVMs inherited the training data bias of MCVSVMs, and total margin of TM-v-SVM was retained, so TM-v-MCVSVMs have the advantages of MCVSVMs and TM-v-SVM, and TM-v-MCVSVMs have better classification performance and anti-noise performance. Then proved it through the noisy face classification experiments of BioID face database.In the fourth section, Total Margin v Minimum Class Variance Support Vector Machines Based on Common Vectors (TM-v-M(CV)2SVMs) was presented. The advantage of TM-v-MCVSVMs was retained on TM-v-M(CV)2SVMs, and the matrix S com of CVs have been introduced to the TM-v-M(CV)2SVMs. The result of noisy face classification experiments of BioID face database shows that TM-v-M(CV)2SVMs have better classification performance and anti-noise performance. In the fifth section, Minimum Class Variance Support Vector Machines Based on Common Vectors (CV-MCVSVM) was presented. CV-MCVSVM based on TM-v-M(CV)2SVMs, and improved the definition of optimization problem. In the new definition, every training data simple minus the mean of CVs, so the same informations of training data were deleted, and more classification informations were retained. Moreover, CV-MCVSVM has the advantages of sparseness, so CV-MCVSVM has faster computing speed on quadratic programming problem. Then proved it through the noisy face classification experiments of BioID face database and my own face database.