An image segmentation method based on a color space distribution model

Summary form only given. The use of image segmentation methods to perform second generation image coding has received considerable research attention because homogenized partial image data can be efficiently coded on a separate basis. Regarding color image coding, conventional segmentation techniques are especially useful when applied to a uniform color space, e.g., Miyahara et al. ( see IEICE Trans. on D-II, vol.J76-D-II, no.5, p.1023-1037, 1993) developed an image segmentation method for still image coding which performs clustering in a uniform color space and implement segment integration techniques. One drawback of such methodology, however, is that the shape of the distribution of color data is considered as a "black box". On the other hand, the distribution of data for an object in a scene can be described by the "dichromatic surface model", where the light, which is reflected from a point on a dielectric nonuniform material, is described by a linear combination of two components, i.e., (1) the light reflected off the material surface, and (2) the light reflected off the inside of the material body. Based on this model, we propose a heuristic model for describing the distribution shape using one or more ellipses corresponding to an object body in uniform color space, where the start and end points of each ellipse are both on the luminance axis. To test the method's performance, we carried out a computer simulation.