Segmenting color images using normalized deviation body reflectance

Reliable feature detection is the basis of image segmentation. Color is the intrinsic feature of the object surface. Hence color could be regarded as the important feature for image segmentation. In the light of this fact, this paper presents a segmentation method of color images, which is based on the normalized deviation body reflectance (NDBR) of the objects. NDBR has the novel feature of resisting the influence of the illumination and the geometry conditions. So far it represents the intrinsic color property of the object, and can be utilized as the reliable evidence for segmentation and recognition. There existed highlights, shading areas, and shadows in the experiment images. Therefore we process these areas respectively through the initial segmentation. Through looking into the properties of the NDBR feature, we find out that the NDBR feature represents hue information of the object. Therefore, we combine saturation and intensity information together with the NDBR feature so as to improve the reliability of segmentation. Also, we present a fast algorithm to extract the NDBR feature from color signals. In addition, in order to improve the reliability of the segmentation, a region-merging technique constrained by a neighboring relation and notable edge information is proposed in our segmentation method. This paper includes experiment results and analysis of the results