Wavelet Transform In Compression Of Still Images

This thesis introduces the application of wavelet transform in image compression,from EZW and SPIHT to JPEG2000 still image compression standard;some wavelet coding techniques are especially introduced:lifting scheme and ROI coding techniques. On the basis of these,we further bring forward a method to decrease the computational complexity of lifting scheme,which is used in JPEG2000 and shows high performance.This dissertation first retrospects the basic principles of JPEG compression standard,and then points out the deficiency of JPEG:square effect when the CR is very high,and only if DCT is taken the place of by DWT,this shortcoming can be overcome and some new features can be realized:such as progressive coding,from lossy coding to loseless coding and ROI coding. In this thesis,we subsequently introduce the basic concepts of wavelet transform,especially the pyramidal algorithm and its fast realization-lifting scheme;this scheme can decrease the time and space complexity of wavelet transform greatly.Wavelet transform,the most perfect mathematical tool of signal processing theory,makes great success in the field of data compression. EZW and SPIHT are the milestones of wavelet image coding on the noiseless channels,and JPEG2000 adopts the EBCOT algorithm. This thesis introduces some wavelet coding techniques of JPEG2000 in detail:such as lifting scheme and ROI coding techniques. As for the high computational complexity of two-dimensional wavelet transform,we propose corresponding simplified method. Taken CDF9/7 filter as an example,this method can decrease the multiplications of lifting scheme by 25%,and can improve the performance considerably. The corresponding paper is presented in the IEEE International conference IXM1C.This thesis not only includes the simple expatiation for the basic principles,but also includes the development of wavelet coding at homeand abroad;and at the same time,we make comparisons among different coding methods and propose our own new idea.