Feature preserving image compression

  • Authors:

  • Kameswara Rao Namuduri;Veeru N. Ramaswamy

  • Affiliations:

  • ECE Department, Wichita State University, 1845 Fairmount Avenue, Kansas, KS;Broadband Engineering, Comcast IP Services, Comcast Corporation, 3 Executive Campus, Fifth Floor, Cherry Hill, NJ

  • Venue:
  • Pattern Recognition Letters
  • Year:
  • 2003

Quantified Score

Hi-index 0.10

Visualization

Abstract

Image details appear as wavelet coefficients with large magnitude in the wavelet transform domain. Image compression methods such as the embedded zerotree wavelet encoding and the set partitioning in hierarchical trees select wavelet coefficients in the order of their significance (magnitude) and encode them generating an embedded bit stream. In existing wavelet based image compression techniques, the significance of a wavelet coefficient is solely defined by its magnitude.In this paper, we describe a flexible scheme to prioritize wavelet coefficients based on the features they exhibit. The proposed scheme combines tree based wavelet coefficient representation with the implicit transmission of data about image features that need to be emphasized. The experimental results presented in this paper demonstrate that it is possible to enhance the image features in the reconstructed images by embedding locally adaptive image processing techniques in the compression algorithm. The main advantage of the proposed technique over the existing methods is that it exploits the embedded zerotree data structure to eliminate the need to send side (additional) information to the decoder regarding the feature selection process.