Novel Multi-Class SVM Algorithm for Multiple Object Recognition

Object recognition is a fundamental task in applications of computer vision, which aims at detecting and locating the interested objects out of the backgrounds in images or videos, and can be originally formulated as a binary classification problem that can be effectively handled by binary SVM. Although the binary technique can be naturally extended to solve the multiple object recognition, which are known as one-vs.-one and one-vs.-all techniques, but the scalability of traditional methods tend to be poor, and limits the wide applications. Inspired by the idea presented by Multi-class Core Vector Machine, we propose a novel Multi-class SVM algorithm, which achieves excellent performance on dealing with multiple object recognition. The simulation results on synthetic numerical data and recognition results on real-world pictures demonstrate the validity of the proposed algorithm.