Error diffusion in Block Truncation Coding

Abstract

A comparative experimental study of several modifications for Block Truncation Coding (BTC) method of gray scale image compression is presented. Two aspects of image processing are considered: blocking and quantization. For blocking three approaches are chosen: standard partition of image array into 4×4 squared blocks, random shifting of partition origin in consecutive lines of 4×4 blocks, and the partition into 16 element blocks along the generalized Hilbert scan of the image. For the quantization standard BTC binarization technique is augmented by the error diffusion in the raster block scan and in the Hilbert scan too. Experiments were conducted on 12 standard images scanned at 300 dpi and visually evaluated on screens with both 100 and 50 pixels per inch. Six quality objective measures were used: mean squared error MSE, percentage of normalized mean squared error NMSE, signal to noise ratio SNR, range to noise ratio SNR', Hosaka plot's mean area AM, and Hosaka plot's standard deviation area AS.