An Improved Multi-Level Edge-Based Stereo Correspondence Technique for Snake Based Object Segmentation

Abstract

Disparity maps generated by stereo correspondence are very useful for stereo object segmentation because based on disparity background clutter can be effectively removed from the image. This enables conventional methods such as snake-based to efficiently detect the object of interest contour. In this research I propose two main enhancements on Alattar’s method first I increased the number of edge levels, and utilized the color information in the matching process. Besides a few minor modifications, these enhancements achieve a more accurate disparity map which eventually helps achieve higher segmentation accuracy by the snake. Experiments were performed in various indoor and outdoor image conditions to evaluate the matching performance of the proposed method compared to the previous work.