可變動區塊之小波轉換影像編碼

Variable Blocksize Wavelet-Transform Image Coding

吳家琪(Jia-Chyi Wu) ； 施銘鎧(Ming-Kai Shih)

《科學與工程技術期刊》 1卷2期 (2005/09) Pp. 57-67

https://doi.org/10.7117/JSET.200509.0057

摘要

小波轉換(wavelet transform)可提供具有對影像多重解析(multi-resolution)的功用，所以被廣泛地使用在影像壓縮技術上。而為了增加影像的壓縮比率，一般將小波轉換後的小波係數經由向量量化(vector quantization)來進行壓縮，而使用向量量化的主要原因是由於它具有很高的壓縮比率，而且在解壓縮時也相當容易。但在實用上，影像壓縮法除了壓縮比率外，編碼速度也是考量的重點。所以在本研究中，我們提出用四分樹分割法作為影像的前處理(preprocessing)，將影像的背景和細節部份分割成尺寸大小不同的區塊，然後經小波轉換後，再將每一個子頻帶區塊的小波係數能量值(variance)做為位元分配的依據。而依照分配到的位元選擇最適當的編碼簿(codebook)，並由編碼後所得到的失真值來調整位元率。這樣的壓縮方式下，在低位元率時依然保有良好的視覺品質，而且由於編碼簿變得較小，所以運算速度也相對的加快許多。本文所提方法的成果與效能及處理時間將於模擬結果處展示。

關鍵字

四分樹分割 ； 小波轉換 ； 位元分配 ； 向量量化

並列摘要

The wavelet-transform procedure has received considerable attention because it provides a multiresolution representation of images. The wavelet transform is now a widely used technique in image compression. By way of a vector quantization technique, the wavelet transformed coefficients can be compressed further. For real-time image compression, achieving a high-data compression ratio and coding efficiency simultaneously are both important. In this study, a quadtree segmentation method for image preprocessing is proposed. A quadtree segmentation algorithm is used to divide a given image, in which regions with image detail will be segmented into blocks with smaller sizes, and the background of the image will be assigned larger block sizes. After the wavelet transform, to select an appropriate size for the codebook, a bit assignment is applied in association with the variance of each subband image block. A rate-distortion concept is also adopted to adjust the compressed bit rate. For this proposed compression scheme, the simulation results indicate that an acceptable visual quality and a high compression ratio can be achieved simultaneously. Furthermore, due to the small size of the codebook, the computational time can be reduced. A system performance analysis is also demonstrated in this study.

並列關鍵字

ouadtree segmentation ； wavelet transform ； bit allocation ； vector quantization