利用線段參數化與混色遮罩輔助之圖像放大

Abstract

隨著顯示裝置的解析度提高，高解析度內容的需求也日漸增加，因此數位影像放大的演算法也越顯重要。在數位影像當中，物體邊緣對於影像品質有很大的影響。因此近代的相關研究皆試圖強化物體邊緣，提升影像品質。 在這篇論文中，我們提出了一套將影像中的物體邊緣形狀參數化，並藉此來保留放大影像中的物體邊緣形狀的方法。因為這些物體的邊緣會混合不同區域的顏色，因此在這個系統中，我們利用混色遮罩找出這些邊緣附近區域的顏色組成，以在放大的圖像中取得該區域的顏色。 最後根據放大圖像中每個像素的位置、該位置附近的物體邊緣以及顏色組成，算出一張具有銳利邊緣的數位影像。

As the resolution of output device increases, the demand of high resolution content has become more and more eagerly. As a result, the image super resolution algorithm becomes more and more important. In digital image, image edge is related to human perception heavily, so image edge is very important to image quality. Because of this, most recent research topics on computer vision that handle digital images do their best to enhance image edge to achieve better quality. In this project, we propose an edge preserving super resolution algorithm, which is related to image vectorization strongly. We first parameterize the image edges to fit edge shape, and than using these data as constraint of super resolution. However, the color nearby edge is usually a combination of two different regions. To get pure color of edge, we use matting technique to solve the problem. Finally we do super resolution based on edge shape, position and nearby color in-formation to compute a digital image with sharp edge.