Compound coding

 -- Mark Kalman, Daniel Wang, Isaac Keslassy

mkalman@stanford.edu
lechuck@stanford.edu
keslassy@stanford.edu

One can often attain improved compression by representing different
elements of an image (such as a background or text) with different
codes.  Consider DCT coding.  Using the same quantization matrix for all
the blocks of an image can be quite adequate for images with uniform
characteristics (photographic only or text only), but it yields poor
results with compound images.  Compound images are those that combine
text, graphics and images (such as a project report!).

Compressing compound images is especially challenging because text and
images have different frequency characteristics and different
sensibilities to noise.  Thus, codes ideally suited for pictures can often
result in text with poor readability.  Conversely, codes well suited for
text or line art yield poor picture quality.  Therefore,  the quality in a
certain region of an image is highly sensitive to the interaction between
the region-classification and quantization parameters.

For our project we will implement and compare four coders: 
1) DCT-based coder, 
2) DCT-based coder with quantization that varies w.r.t region
classification, 
3) Wavelet-based coder,
4) Wavelet-based coder with coefficient quantization that varies w.r.t
   region classification.
 
We plan to compare the performance of the four coders on compound
documents using:

1) PSNR vs. rate.
2) Subjective quality vs rate. 


Main recent reference documents: 

K. Konstantinides and D. Tretter, " A JPEG
Variable Quantization Method for Compound Documents ", IEEE Transactions
on Image Processing, Vol.9, No.7, July 2000, p.1282

J. Li and R. Gray, Text and Picture Segmentation by the Distribution
Analysis of Wavelet Coefficients.  In Proceeding of International
Conference on Image Processing, October 1998.

P. Cosmen, T Frajka, D. Schilling, K. Zeger, "Memory Efficient Quadtree
Wavelet Coding for Compound Images",  Proceedings of the 33rd Asilomar
Conference on Signals, Systems and Computers, Oacific Grove, California,
October 24-27 1999.