A Consistent, Real-Time Image Segmentation for Object Tracking

Abstract

A novel colour image segmentation method designed for region based image registration and video object tracking is proposed. For these applications consistency of segmentation across video frames is very important. To reduce computational complexity, the image is separately segmented in the hue and value feature spaces before combining the results together. Experimental results comparing the proposed and other competing methods are presented and show that the proposed method achieves promising results for natural colour image segmentation in image registration application.