What is it?

EPWIC-1 is a grayscale image compression utility written in C. It is based on a wavelet pyramid decomposition whose coefficients are encoded (one bitplane at a time) using a static arithmetic encoder. The arithmetic encoder uses a generalized Laplacian (or generalized Gaussian) model of the first order statistics of the subbands. This coder was developed in summer 1996, and first presented at ICASSP-97. Compression quality is quite good (see images below), especially considering the simplicity of the algorithm.

EPWIC-2 is a more powerful coder that uses a within-subband and inter-subband conditional statistical model for natural images. Specifically, it estimates the amplitude of wavelet coefficients based on the amplitudes of neighboring coefficients within the subband, and the amplitudes of coefficients in nearby spatial locations in other subbands. Both coders are described in detail in Grasp Laboratory TechReport #414 (May 1997), which was later published in IEEE Trans Image Processing, 8(12):1688-1701, Dec 1999.

C source code for EPWIC-1 is available as a gzipped tar file. The software includes the encoder, the decoder, and a progressive viewer for X windows platforms. Information about the source code as well as copyright information is given in a README file. The latest minor enhancements can be found in the ChangeLog. Source code for EPWIC-2 is not available.

We are making this code available to interested researchers who wish to experiment with a subband pyramid coder. These programs shall not be used, rewritten, or adapted as the basis of a commercial software or hardware product without first obtaining appropriate licenses from University of Pennsylvania.

Example Images Compressed with EPWIC-1

Click to see full size images. (Right click to see the images in a separate browser.)

Orig 256K (8 bpp)	16K (0.5 bpp)	4K (0.125 bpp)	2K (0.031 bpp)
PSNR	35.77 dB	30.03 dB	27.49 dB

Orig 256K (8 bpp)	16K (0.5 bpp)	4K (0.125 bpp)	2K (0.031 bpp)
PSNR	24.92 dB	21.56 dB	20.60 dB

Next page: Example Animated Progression and Results

Last updated by Robert Buccigrossi: 9/19/97