The Influence of DCT Blocksize and Coding Efficiency
Ayodele Embry, embrya@stanford.edu


The fixed blocksize has been a popular techique for exploiting temporal
redundancies in video and minimizing computation and complexity in still
images.  However, as we move towards standards which no longer have a
static data size (i.e. frame) as in MPEG-4, it becomes important once
again to look at the optimal blocksize for coding both still images and
video.  A better approach would allow the motion field to adapt to the
activity in the scene. Specifically, smaller blocks should be used in
areas that are hard to compensate while large blocks are appropriate when
the motion or the scene is homogeneous.  Further, manipulating blocksize
could provide better still image compression since areas of an image which
have lots of detail could take advantage fo the smaller block size for
better resolution.

In this project, I propose to use Matlab and experiment with the relative
gains associated with varying the blocksize for DCT-II.  I will run
experiments on the DCT with blocksizes of 2x2, 4x4, 8x8, 16x16, 32x32, and
64x64.  I will apply both the DCT and IDCT to several images and examine
both the perceptual differences in the decoded image and the resulting
gain in encoding efficiency.  I will also provide incite into why changing
the blocksize affects the resulting gain.