Image compression commonly is achieved using prediction of the value of pixels from surrounding pixels. Normally the choice of pixels used in the prediction is restricted to previously scanned pixels. A better prediction can be achieved if pixels on all sides of the pixel to be predicted are used. A prediction and decoding method is proposed that is independent of scanning order of the image. The decoding process makes use of an iterative decoder. A sequence of images is generated that converges to a final image that is identical to the original image. The theory underlying noncausal prediction and iterative decoding is developed. Convergence properties of the decoding algorithm are studied and conditions for convergence are presented. Distortions to the prediction residual after encoding can be caused by storage requirements, such as quantization and compression and also by errors in transmission. Effects of distortions of the residual on the final decoded image are investigated by introducing several types of distortion of the residual, including (1) alteration of randomly selected bits in the residual, (2) addition of a sinusoidal signal to the residual, (3) quantization of the residual and (4) compression of the residual using lossy Haar wavelet coding. The resulting distortion in the decoded images was generally less for noncausal prediction than for causal prediction, both in terms of PSNR and visual quality. Most noticeably, the streaks found in the decoded Image after causal encoding were absent with noncausal encoding.
Library of Congress Subject Headings
Image transmission; Image compression; Image processing--Digital techniques
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Marchand, J. F., "Image compression using noncausal prediction" (1998). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus