ECE 734: Project

Team members: Jashnani,Yogesh & Mandal,Saikat.

Title: Computationally efficient algorithm for parallel implementation
of zerotree coding

Abstract:

The Encoded Zerotree Wavelet (EZW) algorithm is one of the leading image
compression techniques today.  Current EZW implementations utilize
traditional wavelet transforms followed by zerotree coding.  Although
attempts have been made to improve the performance of the main encoding
algorithm, there has been comparatively little work done towards
reducing its overall complexity. 

We use a separable lifted wavelet transform which leads to significant
reduction in memory requirement.  Compared to the traditional filter
bank implementation, it reduces the number of MAC operations and can be
executed in parallel.  This modification makes our algorithm suitable
for hardware implementation.  Further, we use lifting to build lossless
non-linear integer-to-integer transforms.  This drastically reduces
memory and processing requirements, driving down hardware costs. 

In the main encoding algorithm, we incorporate a fast technique for
identifying zerotrees for all dominant passes in the encoder prior to
the encoding.  This cuts down the encoding time and thus is efficient
for both hardware and software implementation. 

The results of our implementation will be compared with existing SPIHT
and EZW algorithms.  The performance parameters which we will consider
are memory requirement, speed, PSNR etc. 

References:

Shapiro, J.  "A fast technique for identifying zerotrees in EZW
algorithm", Proc.  of IEEE International Conference on Acoustics,
Speech, and Signal Processing, May 1996, Vol.  3, pp.  1455-1458. 

Daubechies, I.  , Sweldens, W.  "Factoring wavelet transforms into
lifting steps," Technical Report, Bell Laboratories, Lucent
Technologies, 1996.