一个使用动态扩张树解决影像压缩问题的分割演算法.doc

一個使用動態擴張樹解決影像壓縮問題的分割演算法 ELSA: A New Image Compression Using An Expanding-Leaf Segmentation Algorithm 研究生：朱浚煌 Ju, Jiun-Huang 指導教授：蔡正發 Tsai, Cheng-Fa 【摘要】 隨著這十幾年來資訊科技的快速進步，許多網路應用服務已變得更加豐富，使用者需求也更加多元化。而這些數位內容造成儲存設備及網路頻寬大量的消耗，特別是多媒體方面的應用。因此，數位內容壓縮在現今的網路環境中就成為一項重要且常見的技術。本論文揭示一種新的影像壓縮技術稱為ELSA。在非頻率領域之有損式影像壓縮技術中最廣為人知的即是LBG Linde, Buzo and Gray ，其優點為速度快及容易實作，但由於其分群過程依靠簡單的距離判斷規則及隨機決定初始編碼簿等特性，使其分群品質不甚良好。而SOM Self-Organizing Map 及HSOM Hierarchical SOM 此種類神經演算法雖然其以相當精細的訓練步驟來得到十分細膩的壓縮結果，但最大的缺點就是速度過慢。而本論文改良LBG的隨機初始化方式，以更有效率的初始化方式來快速得到概略的初始編碼簿，最後再用一正規的LBG對初始編碼簿進行優化步驟。實驗結果顯示，本論文提出的影像壓縮技術在壓縮時間 Codebook Time 及壓縮品質 PSNR 上皆大幅優於其他已發表在國際期刊上之著名演算法，如LBG、SOM、HSOM及INTSOM。 關鍵字：影像壓縮、向量量化、編碼簿、LBG 【Abstract】 The popularity of multimedia on the Internet has created a heavy load on storage capacity and network bandwidth. Consequently, digital content compression techniques have become significant and popular topics recently. Vector Quantization VQ is the most popular method in lossy image compression. An appropriate codebook design is an essential and helpful principle for Vector Quantization such as LBG Linde, Buzo and Gray , SOM Self-Organizing Map and HSOM Hierarchical SOM . The features of LBG are fast and simple, but the compression quality of LBG is not so stable. Even though SOM and HSOM yield the satisfied results, they are too time-consuming. This thesis develops a new image compression method named ELSA, which employs an expanding leaf to determine the rough vectors codewords quickly and utilizes an LBG for quality improvement in the end. Experimental results reveal that ELSA outperforms LBG, SOM, HSOM and INTSOM in terms of computational cost and quality. Keywords：image compression, vector quantization, codebook, LBG