Research On Support Vector Machine And Application On Image Recognition

Support Vector Machine (SVM) is attracting extensive attention as a novel method in data mining. It is based on strict theory and can solve the problem of overcoming local solution and over fitting problem of other algorithms. Also it has very good generalization capacity. As a result, it can be applied in areas such as fingerprint, face recognition, DNA detection, text classification, OCR, handwriting, disease detection and failure detection in auto-controlled equipment.On the base of the statistic leaning theory, this paper investigates on the issues of model construction, fast training algorithms, kernel function construction and regression algorithms of SVM. Then as a validate process, we apply the SVM method to the defects recognition in wood images and the result is positive.The main work of this paper includes:1. A Multi-Lagrange multiplier Support Vector Machine fast training Method (MLSVM) based on the coordinated optimization of multi-Lagrange multipliers is proposed and four individual algorithms, MLSVM1, MLSVM2, MLSVM3 and MLSVM4 are presented. Tested with the standard test data sets of Adult, Web and MNIST, MLSVM3 performs faster than the SMO algorithm with an improvement of 7.4% to 4130% and MLSVM4 performs faster than the SMO with an improvement of 300% to 4200%.2. A novel kernel function that gives attention to the similarity of both input space and feature space is proposed. We also present a novel orthogonal chebyshev kernel function. These functions perform well in tests with test data sets.3. A support vector regression method based on classification is presented to solve the nonlinear regression problem with unknown data distribution and mathematical model.4. The SVM method are applied in identifying internal log defects using CT Imagery and it can automatically find the defects of knot, splits and decay. Next the 3D visual images of log will be reconstructed in computers and the 3D defects will be recognized using SVM.