Segmentation-assisted compression of multimedia documents

Block-based compression algorithms have found widespread use in image and video compression standards. Algorithms such as JPEG, however, while very effective in compressing continuous tone images, do not perform well in compressing multimedia documents which contain text and graphics. With a growing number of applications using images with a high content of both continuous tone data and text, e.g. color facsimile data and educational videos, requires the need for a good compression algorithm. In this paper we present a modified JPEG algorithm for compressing such mixed-mode images which occur in multimedia documents. The algorithm is based on the classification of a block as a kind of text or an image block depending on its DCT coefficients. There can be anywhere from 2 to 256 classes. With such a classification, the same compression algorithm can be applied except that different quantization matrices are used for the different type of blocks. The compression algorithm also uses different entropy codes for the different type of blocks. Simulation results show that the modified JPEG algorithm achieves upto 3 dB gain in PSNR compared to JPEG on multimedia documents.