Combining nonlinear multiresolution system and vector quantization for still image compression

Conference · Fri Dec 17 04:00:00 EST 1993

OSTI ID:10140323

Wong, Y

It is popular to use multiresolution systems for image coding and compression. However, general-purpose techniques such as filter banks and wavelets are linear. While these systems are rigorous, nonlinear features in the signals cannot be utilized in a single entity for compression. Linear filters are known to blur the edges. Thus, the low-resolution images are typically blurred, carrying little information. We propose and demonstrate that edge-preserving filters such as median filters can be used in generating a multiresolution system using the Laplacian pyramid. The signals in the detail images are small and localized to the edge areas. Principal component vector quantization (PCVQ) is used to encode the detail images. PCVQ is a tree-structured VQ which allows fast codebook design and encoding/decoding. In encoding, the quantization error at each level is fed back through the pyramid to the previous level so that ultimately all the error is confined to the first level. With simple coding methods, we demonstrate that images with PSNR 33 dB can be obtained at 0.66 bpp without the use of entropy coding. When the rate is decreased to 0.25 bpp, the PSNR of 30 dB can still be achieved. Combined with an earlier result, our work demonstrate that nonlinear filters can be used for multiresolution systems and image coding.

• Research Organization: Lawrence Livermore National Lab., CA (United States)
• Sponsoring Organization: USDOE, Washington, DC (United States)
• DOE Contract Number: W-7405-ENG-48
• OSTI ID: 10140323
• Report Number(s): UCRL-JC--115838; CONF-940247--3; ON: DE94009422
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200
ALGORITHMS
COMPRESSION
IMAGE PROCESSING
MATHEMATICS AND COMPUTERS
PATTERN RECOGNITION
VECTOR PROCESSING