利用高曲率特徵點之影像壓縮技術

Abstract

在本文中，我們利用影像曲面上之高曲率特徵點之特性，設計一具有漸進式傳輸能力的影像壓縮架構。透過高曲率點萃取與連結的動作，我們首先可以得到影像資料的「崖角」特徵；同時在此過程中，我們使用各種不同的排列和搜尋的演算法，標記這些影像崖角的重要性，以作為資料壓縮之漸近傳輸與適應性量化的依據。此外，我們採取兩種不同的影像崖角表示 法：直接座標記錄以及 B-spline 最小均方差逼近法，並相對應地選用適當的編碼技巧，（如線性預測編碼）以去除崖角特徵點之間的累贅資訊。最後我們使用熵編碼法來進一步精簡影像編碼資料。經由實驗與討論，我們相信一旦在特徵抽取及其處理的技術上能有所突破，本文所提出的架構可使用於高倍率之影像壓縮。

In this thesis, we propose an image compression scheme based on the high-curvature features of image surfaces. First, we apply Jong-Wang operator to extract the desired features and link them together into feature curves, which are named as "image brinks" in this thesis. Then we rearrange these brink features in order of "importance", to facilitate progressive data transmission as well as adaptive quantization according to the significance of an image brink. Moreover, we represent the image brinks by directly keeping their 3D coordinates or by using B-spline approximation. With different representation methods, various strategies for compact coding, like the linear predictive coding, may be adopted. In the last stage, entropy coding is performed to further reduce the coding redundancy. Through the observation and discussion in this thesis, we believe that the proposed coding scheme is suitable for high-ratio compression, provided that there is sufficient progress in feature extraction and manipulation algorithms.