Seminar: Competing One-class Support Vector Machines

Yiming Qian
M.Sc. Thesis Proposal
Supervisor: Dr. Minglun Gong

Competing One-class Support Vector Machines

Department of Computer Science
Friday, October 4, 2013, 11:10 a.m., Room EN 2022


The Support Vector Machine (SVM) is one of the popular batch learning methods for classification due to its high accuracy. However, for binary classification, binary SVM algorithm only train one SVM model depicting one separation hyperplane, which may not provide sufficient confidence on labeling ambiguous data. This talk presents a SVM classifier towards the drawback of SVM. The key idea is to solve both binary and multi-class classification problems using multiple competing one-class support vector machines (C-1SVMs). It utilizes the advantage of online learning, producing a partially trained model immediately, which is then gradually refined toward the final solution. It can also achieve faster convergence. Real-world problems of foreground segmentation and boundary matting for live videos in Computer Vision have been solved to demonstrate the effectiveness of C-1SVMs. The results are shown to be particularly competent at processing a wide range of videos with complex backgrounds from freely moving cameras.