Improved Mixed Document Compression by Using the DCT Coefficient Distributions

Abstract

Data compression is an important area in electronic document processing. A document can consist of text and image, with different statistical behavior. When we use JPEG compression, it is advantageous to tune the parameters for individual blocks to enhance the decompression quality. We demonstrate a method that can be incorporated in the JPEG decoder, which utilizes the amplitude distribution of the DCT coefficients for texts and images to achieve better image quality. This involves both a discriminator function to differentiate between text and image, and an adjustment to the decompression methodology that shifts the decoding value to the minimum mean-square error location in the codeblock.