Region segmentation coding of color images based on uniform hvc color space

There are two aspects to the region segmentation coding and the model‐based coding which extract and encode the structure and features from image information. They are the “technique for compression of image information” and the “technique for structural description of image information.” If the image information is described structurally, not only is the highly efficient coding realized but also a new image can be constructed by a local manipulation of the original image. This paper combines the region segmentation of the image based on Munsell color space and the region merge based on the hue difference between the regions. Then a new region segmentation coding based on the result of the forementioned region segmentation is proposed. It is assumed that an object consisting of the color image corresponds to a region of almost the same hue. The hue information within the region is coarse‐quantized based on the memory colors, and the shape information is coded using the chain code. For the value information which contains the information for the shape, shade, etc., of the object, the adaptive cosine transform coding is applied. For the chrome information, the region is approximated by a polynomial function and its coefficients are encoded. The proposed coding method is applied to several standard images which are portraits of relatively simple structures with different face sizes. The color information is transmitted with a relatively high efficiency, occupying approximately one‐fourth of all the transmitted information; and the usefulness of the region segmentation proposed in this paper is demonstrated. Compared to the traditional waveform coding, the method has the advantage that the local manipulation of the image is easier, although the total transmitted information is slightly larger.