Significance-linked connected component analysis and morpho-subband decomposition for multiresolution image compression

Image compression has become a topic of increasing importance as image processing systems and applications come of age. Continuing cost improvements in computing power, storage and communications are making such systems more practical, with compression almost always included to achieve cost-effective solutions.; Subband decomposition of an image achieves energy compaction, and spatial-frequency localization. It leads to the development of a new generation of image compression techniques. Subband (wavelet) image coding, also known as multiresolution image coding, has been an active research topic over the past decade. Recent success in wavelet coding can be mainly attributed to recognition of the importance of data organization and representation strategies. These strategies improve the PSNR of wavelet coders by 1 to 3 dB over block-based image coders, such as JPEG.; In this dissertation, existing image compression techniques are reviewed, and limitations with existing algorithms are discussed. Two new techniques for multiresolution image compression are then proposed. The first technique is a novel data representation algorithm termed significance-linked connected component analysis (SLCCA) for wavelet image coding. There are four key features in the proposed SLCCA, they are: multiresolution discrete wavelet image decomposition; connected component analysis within subbands; significance-link registration across subbands; and bit-plane encoding of magnitudes of significant coefficients by adaptive arithmetic coding. Two image coding algorithms have been constructed based on SLCCA, one for lossy compression, the other for lossless compression. Extensive experiments indicate that the proposed algorithm outperforms published wavelet coding algorithms for lossy compression. The lossless algorithm outperforms the international standard--lossless JPEG, and most of the published subband lossless compression techniques.; The second technique, 1-D morpho-subband image decomposition, aims at reducing the visually annoying "ringing effect" present in linear-filter-based lossy coding algorithms at very high compression. By using a morphological filter in place of a linear filter around edges in an image, the ringing effect is greatly reduced compared to the linear-filter-based wavelet coder. In addition, 1-D morpho-subband image compression outperforms other morphological subband coders by 2-3 dB in PSNR and has greatly improved visual quality.