Trellis coded quantization has recently evolved as a powerful quantization technique in the world of lossy image compression. The aim of this thesis is to investigate the potential of trellis coded quantization in conjunction with two of the most popular image transforms today; the discrete cosine transform and the discrete wavelet trans form. Trellis coded quantization is compared with traditional scalar quantization. The 4-state and the 8-state trellis coded quantizers are compared in an attempt to come up with a quantifiable difference in their performances. The use of pdf-optimized quantizers for trellis coded quantization is also studied. Results for the simulations performed on two gray-scale images at an uncoded bit rate of 0.48 bits/pixel are presented by way of reconstructed images and the respective peak signal-to-noise ratios. It is evident from the results obtained that trellis coded quantization outperforms scalar quantization in both the discrete cosine transform and the discrete wavelet transform domains. The reconstructed images suggest that there does not seem to be any considerable gain in going from a 4-state to a 8-state trellis coded quantizer. Results also suggest that considerable gain can be had by employing pdf-optimized quantizers for trellis coded quantization instead of uniform quantizers.
Library of Congress Subject Headings
Image compression--Research; Coding theory--Research; Vector processing (Computer science)--Research
- Please Select One -
Department, Program, or Center
Electrical Engineering (KGCOE)
Panchapakesan, Kannan, "A Study of trellis coded quantization for image compression" (1996). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus