本文為澳大利亞莫納什大學（作者：Torsten Seemann）的博士論文，共300頁。

本文提出了一種實現“區域性分割”的低層次影象處理的統一思想。區域性分割提供了一種檢查並理解現有演算法的方式，以及建立新演算法的範例。區域性分割可以應用於重要的影象處理任務。利用傳統的強度閾值分割技術和簡單的模型選擇準則，與現有演算法相比，新的FUELS去噪演算法顯示出很強的競爭力。為了改進區域性分割，採用最小訊息長度的資訊理論模型選擇準則（MML）對不同結構和複雜度的模型進行選擇，從而進一步提升了去噪效能。FUELS和MML變體演算法不需要特殊的使用者提供的引數，而是從影象本身中進行學習。我們相信，影象處理可以從區域性分割方法的應用中受益匪淺。

A unifying philosophy for carrying out low level image processingcalled “local segmentation” is presented. Local segmentation provides a way toexamine and understand existing algorithms, as well as a paradigm for creatingnew ones. Local segmentation may be applied to range of important imageprocessing tasks. Using a traditional segmentation technique in intensitythresholding and a simple model selection criterion, the new FUELS denoising algorithmis shown to be highly competitive with state-of-the-art algorithms on a rangeof images. In an effort to improve the local segmentation, the minimum messagelength information theoretic criterion for model selection (MML) is used toselect between models having different structure and complexity. This leads tofurther improvements in denoising performance. Both FUELS and the MML variantsthereof require no special user supplied parameters, but instead learn from theimage itself. It is believed that image processing in general could benefitgreatly from the application of the local segmentation methodology.

1 引言
2 符號與術語
3 影象處理中的區域性分割
4 區域性分割去噪
5 基於資訊理論的區域性分割
6 區域性分割的擴充套件與進一步應用
7 結論

下載英文原文地址：

http://page5.dfpan.com/fs/6lcj32219291561e1b2/

更多精彩文章請關注微訊號：