Bandelet based vector quantization coder design for gray scale image compression

Abstract

Providing efficient transform-based representations of images is an important problem in the area of image compression. In this paper, the geometrical flow of an image is analyzed by using the Bandelet transform and a new Bandelet based image coding scheme is proposed. First, the bandelet transform of the image is computed. The motivation behind the usage of the bandelet transform is that the geometry of the image is summarized with local clustering of similar geometric vectors, the homogeneous areas being taken from the quad tree structure. This will allow us to search for areas in the image that are geometrically similar to each other. Then the spatial and geometric interpixel redundancies present in the bandelet transformed coefficients are removed. The psycho-visual redundancies are removed using simple vector quantization (VQ) process. Finally, the consequential coefficients are encoded using Huffman encoder. Our experiments demonstrate that the proposed scheme achieves near-optimal rate-distortion performance for natural images. It is reported that a gain in the bit-rate of about 0.83 bpp over the wavelet based algorithms is achieved yielding similar quality factor. Key words: N relations join, query optimization, distributed database, rank aware query, top K.