Development of Efficient Algorithms for JPEG2000 Image Coder

Abstract

Compression of digital images has been a topic of research for many years and a number of image compression standards have been developed for different applications. The role of compression is to reduce bandwidth requirements for transmission and memory requirements for storage of all forms of data. Although, developments on technology front provide high speed digital communication and large memories, image compression is still of major importance. The advances in technologies generally result into increased demand for image communication, as well as demand for higher quality image printing and display. The work presented in this thesis revolves around the improvements in the JPEG2000 image compression standard. We have proposed an Edge Adaptive Wavelet Lifting Scheme that deals with the edges present in an image efficiently. In this approach, the lifting direction is not restricted to horizontal and/or vertical direction only, but other lifting directions are also used if the edges present in an image are neither in horizontal nor in vertical direction. This approach represents the decomposed image more efficiently because there are less number of wavelet coefficients having large magnitude in the high pass subbands. Due to this approach, the energy of high pass subbands decreases and these subbands are represented more efficiently. It has been observed that proposed scheme improves PSNR by more than 4 dB when compared with the JPEG2000 lifting scheme. Tile boundary artifacts in JPEG2000 compressed images have also been discussed in this thesis. We have proposed a post-processing technique to reduce tile boundary artifacts in JPEG2000 compressed images in this work. This technique reduces tile boundary artifacts which occur in an image when it is compressed at low bit rate using JPEG2000 standard. We have analyzed the effect of quantization on the region of the tile boundaries of JPEG2000 compressed images. The analysis confirms that tiling artifacts are reduced by updating the high pass reconstructed samples lying on the boundary of the image tiles where the artifacts occur. Post-processing is applied on the output of JPEG2000 coding system and thus it can easily be blended with JPEG2000 standard. It has been shown that when one uses proposed method, tile boundary artifacts are reduced in terms of increase in PSNR value. This increase in PSNR value varies from 0.1 dB to 1 dB for different images considered in this work. The reductions in boundary artifacts have also been observed in visual quality of decompressed images. Bit rate allocation to the tiles of an image has also been analyzed in this work. In JPEG2000 encoder, equal bit rate is assigned to each tile of an image. This assignment is suitable for the images with information contents equally distributed throughout the image. However, tiles of an image may have different complexities. Some of the tiles may have larger texture area while others may have larger smooth area. The quality of a reconstructed image varies a lot if all tiles in the image do not have same complexity. As such, we should include the complexity of a tile while assigning a bit rate to it. This is a known fact that entropy of a complex tile is more than the entropy of a smooth tile. So, we have proposed a method to assign bit rate to a tile based on the weights derived using zero-order entropy of a tile. Using this method, tiles of an image have been assigned different compression bit rates. Visual quality of reconstructed image is improved by assigning these bit rates to the tiles of an image. The proposed methodology improves the PSNR values for all images and for all bit rates considered in this work. Maximum PSNR improvement by the proposed method is 2 dB. Due to this improvement, the proposed methodology provides better visual quality in JPEG2000 reconstructed images than the conventional approach of JPEG2000 standard. This improvement has also been shown taking place when compared with JPEG2000 encoder and also with other existing algorithm proposed. Next, Post Compression Rate Distortion (PCRD) optimization scheme of JPEG2000 has been discussed in this thesis. It has been observed that a considerable amount of computation and memory usage in PCRD optimization scheme in JPEG2000 is redundant. Motivated by this observation, an efficient rate control approach for JPEG2000 encoder is proposed, in which optimal Rate Distortion (RD) slope threshold is selected using the minimum RD slope of lower level horizontally and vertically low pass subbands of the wavelet decomposed image. Also, the passes of the code blocks of other subbands, whose RD slopes are less than the optimal RD slope, are skipped. The proposed approach reduces memory requirements and encoding time when compared with existing rate control approaches for JPEG2000 standard and also provides the same image quality as that of PCRD approach used in the JPEG2000 standard. The encoding time and memory usage in the proposed approach is 50-60% less than PCRD approach and comparable with existing approaches. The work presented in this thesis, when combined with JPEG2000 standard, shall result into the reduction of storage and transmission costs of digital images.