A new semi-supervised support vector machine classifier based on wavelet transform and its application in the iris image recognition

Abstract

This paper presents a new image recognition framework based on semi-supervised learning and wavelet transform. We first investigates three feature extraction methods based on wavelet transform to extract image features, and then an important semi-supervised learning technology, semi-supervised support vector machine(S3VM), is proposed based on wavelet analysis. However, the resulting problem is a nonconvex and nonsmooth optimization. Furthermore, the problem is posed as a mixed integer programming and thus its global solution is obtained. Moreover, the proposed wavelet-based S3VM is directly applied to recognition iris image dataset. Compared to the corresponding support vector machine (SVM), experimental results indicate the feasibility and effectiveness of the proposed framework when insufficient training information is available.This paper presents a new image recognition framework based on semi-supervised learning and wavelet transform. We first investigates three feature extraction methods based on wavelet transform to extract image features, and then an important semi-supervised learning technology, semi-supervised support vector machine(S3VM), is proposed based on wavelet analysis. However, the resulting problem is a nonconvex and nonsmooth optimization. Furthermore, the problem is posed as a mixed integer programming and thus its global solution is obtained. Moreover, the proposed wavelet-based S3VM is directly applied to recognition iris image dataset. Compared to the corresponding support vector machine (SVM), experimental results indicate the feasibility and effectiveness of the proposed framework when insufficient training information is available.