This project proposes a new algorithm for the distrubuted source coding of
highly correlated images. The algorithm is based on the theorem established
by Slepian and Wolf and the practical implementations of the theorem by
Pradhan. It is similar to Pradhan's work in that it involves the concept
of coset construction and transmission to eliminate the transmission of
information already inherent in the other image at the decoder. The algorithm
deviates; however, from Pradhan's work in that the correlation of data (in
this case still images) is defined by the proximity of the gray level values
of the individual same-location pixels of the image pairs rather than the
Hamming distance proximity of data. This difference is due to that in still
image transmission, clearly, the adjacent pixels are close to each other in
their actual gray level values, and making use of this intra-pixel correlation
is very effective in lowering the overall bit rate. Another difference between
the algorithm proposed here and Pradhan's work is that the coset construction
is done by modulus arithmetic rather than through the use of error-correcting
codes. This is because error-correcting codes are very efficient to construct
cosets with elements differing by large Hamming distances; while modulus
arithmetic suits well for coset construction for cosets with elements
differing by large gray level differences.
It should be noted that the proposed algorithm in this project works very well
for small D values as explained. As for larger D values, the algorithm is
still successful for images with high intra-pixel correlation, while deviations
from the Slepian-Wolf bound occur for images with steep gray level variations
in horizontal scan. However, even in these cases, the algorithm does much
better compared to the individual coding of the same images.
Further research in the algorithm for still images may include combining this
algorithm with an edge detection/quadrature decomposition algorithm to
define the edges in the image and modify the algorithm for the edged areas.
ABSTRACT
INTRODUCTION
PROBLEM DESCRIPTION and PRIOR WORK
DATA SET
ALGORITHM
RESULTS
CONCLUSIONS
REFERENCES