- 學位論文
使用硬體加速於中國水墨畫之白描技法模擬
Applying GPU acceleration to Chinese outline drawing
摘要
本論文探討使用硬體加速模擬中國水墨畫之白描技法。在輪廓線搜尋的部份我們應用兩道著色(shader)流程來改良McGuire and Hughes[1]所提出的邊線網格的方法。第一道流程，我們使用繪圖晶片(GPU)中片段著色器(Fragment processor)來計算非相片寫實繪圖(Non-Photorealistic Rendering)中最重要的輪廓線搜尋，增進運算速度。第二道流程則是連結輪廓線並加上水墨畫筆觸效果。McGuire and Hughes[1]所提出的邊線網格(Edge Mesh)的方法是將原本CPU 的運算交給繪圖晶片(GPU)中頂點著色器(Vertex processor)的處理；在這裡，我們提出的方法則是應用片段著色器的運算，因為片段著色器是屬於單一指令多資料流(SIMD，Single Instruction, Multiple Data)的架構，所以運算速度比頂點著色器更快。在筆觸模擬的部分，因為繪圖晶片的硬體限制，無法獲得整條相連輪廓線段的資訊，因此我們將第一道著色流程處理後的輪廓線資訊存回貼圖中，利用CPU 讀取貼圖找出輪廓線段位置的資訊。最後我們調整四邊形大小產生類似筆觸粗細的變化，再利用畫有水墨筆觸的貼圖呈現出最後的水墨風格。
並列摘要
This thesis addresses the problems of Chinese outline drawing using GPU acceleration. For the feature edge search, we use two shader passes to improve the method of edge mesh proposed by McGuire and Hughe. In the first pass, we use Fragment processor of GPU to compute the important silhouette search of NPR，and it will increase efficiency. In the second pass, linking all silhouette and add Chinese outline drawing. The method of edge mesh proposed by McGuire and Hughe is based on Vertex processor of GPU computation that original using CPU-based computation. The method which we proposed here is based on Fragment processor of GPU computation. Fragment processor is a SIMD (Single Instruction, Multiple Data) architecture, and it’s speed will be faster than Vertex processor. For the brush stroke simulation, the GPU-based method can not obtain whole information of the feature edges due to some restrictions on GPU. We store the silhouettes which computed by first pass to texture, and read texture by CPU to find the location of silhouettes. Finally we resize the quads to produce the thick and thin brush, and use the texture with Chinese Ink stroke to produce Chinese outline drawing.