VECTOR QUANTIZATION OF OPTIMALLY GROUPED SETS AND IMAGE/SPEECH COMPRESSION.

Vector quantization (VQ) of topological sets whose elements are optimally selected is presented. The method includes conventional VQ as a special case. First, this algorithm is explained without introducing any physical entity to the data to be processed. Therefore, the method is applicable to a wide class of data such as image and speech. Then, the given algorithm is interpreted by using the image-coding concept and terminologies. In this case, the whole image is subdivided into convex polygons, e. g. , convex quadrilaterals. The shape of this region is decided by the optimization to a given set of regular polygons. Various problems peculiar to image data are pointed out and discussed. Encoding (image compression) and decoding (image reconstruction) also include the region optimization. This means that the presented method generates side information of the region pattern. However, the increase of the total information can be cancelled out since each region size can be set larger. The case of speech is also given briefly. Using the VQ method, the author develops a class of intelligent pattern handling that reflects the geometry of the source data. Subdivision of the given image/speech is one example.