SPIHT演算法影像編碼加入視覺模組與變換大小

Abstract

SPIHT演算法是近來被廣泛討論的影像壓縮方法。它利用小波轉換(Wavelet transform)加上樹狀結構(Tree structure)很有效率地編碼。在本文中，我們將測試並討論SPIHT演算法。 在改良的部分，由於處理後的圖像是讓人觀賞的，而人對於圖像在不同的頻率有不同的敏感度，所以加入視覺模組(Perceptual model)可以根據人的視覺有效分配資料量，如此有助於在低資料量時有較舒服的視覺效果。 在應用的部分，我們加入了大小變化(Spatial scalable)的彈性，可以應用在瀏覽、下載上。比較小的圖需要的資料量也會相對降低，處理的時間也會縮短。特別的是，我們這個做法在傳輸非原圖大小的資料時只需讀檔頭便可於壓縮後的資料中抓取需要的部分而不需做部分解碼的動作。

The zerotree structure based on wavelet transform was first introduced by Shapiro in his work of embedded zerotree wavelet(EZW) and later improved by Said and Pearlman in their work of set partitioning in hierarchical trees (SPIHT) \cite{SPIHT}. This technique not only was competitive in performance with the most complex techniques, but was extremely fast in execution and produced an embedded bit stream. We develop 2 SPIHT based image coders: 1) image coder incorporating perceptual models, 2) image coder with dual SNR and spatial scalabilities. The perceptual models are used to improve the perceptual quality at the same bit rate while scalability in spatial resolution ensures us the application of transmitting or browsing images of different sizes. 1.1 Outline……………………………………………………………………2 2. SPIHT algorithm …………………………………………………………4 2.1 Wavelet transform and filter banks ………………………………4 2.2 SPIHT coding algorithm ………………………………………………8 2.2.1 Set partitioning……………………………………………………10 2.2.2 Spatial orientation trees ………………………………………10 2.2.3 Coding algorithm……………………………………………………12 2.3 Arithmetic coding ……………………………………………………14 2.3.1 Application of arithmetic coding on SPIHT algorithm ……16 2.4 Experimental results…………………………………………………17 2.4.1 Tests on wavelet……………………………………………………17 2.4.2 Tests on SPIHT………………………………………………………20 2.4.3 Tests on arithmetic coding………………………………………23 3.Wavelet coding incorporating perceptual models…………………27 3.1 Previously developed perceptual models…………………………30 3.1.1 JPEG luminance quantization table ……………………………30 3.1.2 Perceptual weights suggested by L Hontsch, L. J. Karam, and R. J. Safranek…………………………………………………………32 3.2 Perceptual weights calculated from the MTF function ………33 3.3 Experimental results and discussion ……………………………36 3.3.1 Experimental results………………………………………………36 3.3.2 Discussion……………………………………………………………39 4. Scalability in SNR and spatial resolutions ……………………50 4.1 SPIHT based coder with dual scalabilities ……………………51 4.2 Coding example…………………………………………………………54 5. Conclusion ………………………………………………………………57