MY WAVELET COMPRESSION PAGE

This demo is based on the paper "An Image Multiresolution Representation for Lossless and Lossy Image Compression" by A.Said and W.A.Pearlman. Basically, I used the same S+P transform proposed in the article but with a 3-symbol partition set. I then arithmetic-encoded each bitplane (8 in this demo for sake of simplicity) using a semi-adaptive arithmetic encoding (two-pass). I also implemented a version fully-adaptive with 25 different contexts but it's much more complicated and slower. An intoduction to Arithmetic Coding is given here (not the fastest link on the web...).

The property of this methods are:

• Prograssive Transmission: by encoding each bitplane separately it is possible to stop the decoding process at any bitrate and still be able to reconstract the image at the chosen quality.
• Embedded: it is possible to fix at the encoder side how much to compress the image for different applications.
• Lossless/Lossy: by encoding (and decoding) all bitplanes, lossless compression can be achieved. To do so integer resolution transformation is used which solves the problem of finite-precision.
• Fast: the wavelet transform is implemented multiplication-free by bit-shifting.

The alghoritm proposed in the paper works well and I managed to reach around 4.51 bpp lossless for Lena (512x512) with the simplest version. It isn't fully-embedded but it is possible to perform lossy compression by encoding fewer bitplanes (max 15 bitplanes in my implementation for lossless compression). I found better performances (for lossy compression) by quantizing each scale with a different visual mask than by shifting each bitplanes accordingly.

Many other information about the work of these 2 guys can be found in the SPIHT's home page (including lots of papers and cool links).

Cool links for Data compression and Signal & Image Processing

SIPID

Signal Processing and the Multimedia Information Infrastructure

JPEG links

GTI

This gif-animated demo has been created using the gifmerge program (zipped file). It is in C code and quite straight forward to compile. I've only tested it under Unix machines but it should work fine in a PC environment too!

H263+ decoder and encoder

H263dec.zip: C code to decode H263/H263+ bitstream.

h263play.zip: Windows H263 player.

tmn-3_2_0.tgz: Telenor H263 decoder.

tmndec-3_2_0.tgz: Telenor H263 encoder.

H263.zip: All the above software plus other versions.