Segmentation by Fusion of Features in Multiple Color Spaces and Texture Features Based on PRI

For natural image segmentation, due to features from a single image are hard to describe the complex scene information, this paper presents a new method based on the fusion model evaluation index PRI to fuse color histogram features in 3 color spaces, RGB, XYZ, LUV, and texture features. We experiment on images from Berkeley segmentation databases and compare the quantitative and qualitative experimental results with manual segmentation and some classic segmentation methods, such as Mean-shift, FCR, etc. Experimental results show that the results of this paper are more similar to the real segmentation results of manual segmentations. The method proposed by this paper has obvious advantages in solving the contradiction between segmentation accuracy and robustness, and the contradiction between over-segmentation and insufficient segmentation.