Segmentation of Full Vision Images Based on Region Over-Segmentation and Merging

The problem of how to accurately segment the full vision images caught by robots is investigated. First over-segmentation of full vision images based on color features in HSV color space is performed. Then by taking into account the fact that texture features are not sensitive to illumination changes, the regions which have been over-segmented are merged after the texture features are extracted via Fourier transform. It is demonstrated that the proposed method can achieve full-image segmentation and is of higher precision and reliability than simply using the color or texture features.