Resolution scalable and random access decodable image coding with low time complexity
William A. Pearlman, Daniel Z. Freedman, Yushin Cho · 2005
Modern wavelet-based image compression methods provide not only higher compression performance, but also the capability to support various features, such as quality (SNR) scalability, resolution scalability, and region-of-interest encoding and decoding. Quality scalability is commonly achieved via bit-plane coding, which also helps to improve compression, since neighboring bits provide convenient and powerful contexts for entropy coding. However, on many important applications (e.g. digital camera), the images always need to have a pre-defined high quality, and any extra effort required for quality scalability is wasted. Furthermore, for compressing a very large size image source, a low time complexity is often the most desirable characteristic of an image coding algorithm. In this thesis, a resolution scalable and random accessible image coding algorithm, PROGRES (Progressive Resolution Decompression), is designed based on predictive dynamic range coding of wavelet coefficients and without bit-plane coding. Avoiding bit-plane coding leads to considerable speed improvement without compromising coding efficiency. The algorithm is designed and implemented for both 2D and 3D image sources. Experiments show that our suggested coding model lessens the computational burden of bit-plane based image coding, both in encoding and decoding time. The PROGRES algorithm combined with the presented fast random access decoding method having O(log2 n) block seek time is suitable for browsing a very large image bitstream. It can seek the requested part in the code-stream very quickly, and then decode them upto desired resolution at high speed. In related work, we introduce the concept of higher order zerotrees in modern wavelet-based coders and quantify their relative coding power. By analyzing two famous zerotree-based image coders, EZW and SPIHT, we are able to explain the superior coding efficiency of SPIHT through its ability to code higher order zerotrees than EZW. We are also able to calculate the bit savings of SPIHT compared to EZW within this framework.