基於次像素邊緣圖和區域關連性的影像放大方法

Abstract

由於影像的內容是千變萬化的，若利用固定的演算法來對整張影像做放大，便有可能造成影像中的一些部分有著不錯的放大效果，但某些部份則否。本論文提出一種利用次像素邊緣圖和區域關連性做為依據來設計的模糊推論影像放大系統，針對每個小區域的影像特性做出不同的放大方法，以改善輸出影像之品質。 利用基於次像素邊緣圖的內插法可以彌補內插時遺失高頻資訊的缺點。另外，找出影像中的區域關連性可以使我們更加瞭解影像之特性，並且對次像素邊緣圖可能發生的一些錯誤做出修改。在得出次像素邊緣圖和區域關連性資訊後，利用模糊推論系統可以對這些資訊做一個有效率的分析，並依據此分析結果選擇最適合此區塊的放大方法。 在本文中做了一些真實影像的實驗，由實驗中可以發現所提方法可以得出一個頗為自然且符合人眼感覺的輸出影像。

There are miscellaneous kinds of contents in an image. When images are enlarged by an immobile algorithm, the output image quality maybe good in some areas but faulty in certain areas. In this thesis, a method of image magnification is proposed to improve the output image quality by applying the fuzzy logic control system to analyze the information of sub-pixel edge map and local correlation characteristics to infer the appropriate way for enlargement of the small cells of the input image. The edge map-based interpolation method can preserve high frequency information during the process of image magnification. Besides, the image characteristics can be known by using the local correlation information that can be used to modify mistakes in the sub-pixel edge map. After adequate information is available by some steps, the fuzzy inference system can then be used to make information more effective for analysis. And an appropriate magnification algorithm will be chosen according to the inference system outputs. In this thesis, several real images are experimented as inputs of the proposed image magnification system. From these experiments, output images can be obtained with more natural appearance and more consistency with human vision.