採用多重解析度加強邊角之技術

Abstract

在這篇論文中， 我們提出一個新的想法： 利用多重解析度之技術加強模糊影像中之邊角。對一個訊號而言，在不同解析度下許多重要 的資訊仍會被保留下來。至少，在不同解析度下的波形仍具有相當的相似性。我們試著利用在不同 解析度下影像內容變化的趨勢，來估算模糊影像中被濾掉的高頻成份。而根據這些趨勢， 我們可以大略估算這失去的高頻成份。為了使得還原影像的邊角斜率較大、而且振幅波動 較小，在這篇論文中， 我們提出一些方法來改善還原影像的品質， 計有遞迴法， 外差法、和區域性反濾波法。除此之外，我們還探討了事先知道退化機制與否所造成的影響。 基於這樣的想法，模糊影像中大部分型態的邊角，都可能被還原成較接近原始影像之邊角。

In this thesis, we propose a new concept to enhance edges in a blurred image basing on the multiscale analysis. For a signal, some important information has been kept in different scales. Basically, the shapes of a signal in different scales are fairly similar. We try to take advantage of these tendencies to estimate the lost high-frequency components in the blurred edge. According to these tendencies, we may reasonably estimate the lost high frequency part of the blurred image. To have sharper edges and less fluctuation in the restored image, we propose several estimation methods: the iterative method, the extrapolation method, and the windowed inverse filter, to improve the quality of the restored image. Besides, we also discuss the influence of the degradation model. Based on our approach, many types of edges in the blurred image may be enhanced to be closer to their original shapes.