以影像為基礎之棉紙撕畫自動生成系統

Abstract

撕畫是一種將多張紙張分割之後重新拼貼起來的繪畫創作方式，主要風行於亞洲地區，利用手工棉紙豐富的色彩以及細長纖維的特色來創作出獨特的藝術風格。過去大部分非擬真成像的研究，都是集中在用物理模擬或是其他方式來重現出各種筆刷工具的筆觸風格，而本篇論文的重點在於如何呈現出棉紙撕畫邊緣的纖維以及紙張的層次感，讓使用者能夠輕易製作出精美的棉紙撕畫。本篇研究提出了一個方法，讓使用者輸入一張圖片，接著將圖片按照細節程度分成多種不同的層次，再逐一將各個層次的區塊用紋理轉換的技術來模擬對應的材質，並對每個區塊的邊緣按照紋理的方向來模擬纖維毛後，貼到畫布上，最後就能得到一張棉紙風格的撕畫成果圖。

Fabric paper painting is a unique art style in Asia. These works are made from assemblage of different papers. Due to the variation of paper and the feature of fabric, we can create colorful and vivid paintings. Most of the previous NPR studies focus on simulating the style of specific brush tool in physically-based or other methods. In this thesis, we dedicate on presenting the layers of paper and edges of fabric paper. We propose a system, which segments input image into different layers and fragments by mean-shift segmentation, and then simulate the fabric edge according to the image tangent flow. For each layer, we use the technique of texture transfer to mimic the corresponding feature. Finally, we combine all the layers to get a fabric paper style image.