影像風格化是指將輸入影像轉換成具備特殊風格的影像，此轉換過程通常會運用數個特定的圖案樣式來增強影像的特徵。一個好的風格化影像關鍵在於風格圖樣的設計和排列。然而對於初學者而言，設計一個新的風格化圖案並將之適當的排列是不容易的。在本研究中，我們提出一種新的技術，基於非等向性反應擴散(anisotropic reaction diffusion)的影像風格化系統，使圖樣產生及設計風格化影像更為容易。本系統是基於反應擴散來產生圖樣。為了擴展反應擴散系統能產生的圖樣，我們提出的方法包含使用一些調整後的非等向性擴散來將原有之圖樣進行變形並結合流場來引導圖案的排列。亦提供一個圖樣選擇器，以方便使用者從各式圖樣中做選擇。而在後處理步驟中，我們使用一個新的閾值處理和顏色匹配方法，以便於針對影像做尺寸，密度和圖案顏色的改進。從實驗結果和使用者研究的回饋中，我們所提出的多種影像風格化樣式，包括剪紙、風格化半色調和運動錯覺(流場視覺化)影像，均可透過我們的方法產生並且被多數使用者接受，證明此一系統的可行性和靈活性。 Image stylization refers to the process of converting input images to a specific representation that enhances image content using several designed patterns. The critical steps to a successful image stylization are the design of patterns and arrangements. However, only skilled artists master such tasks because these tasks are challenging for most users. In this paper, a novel image stylization system based on anisotropic reaction diffusion is proposed to facilitate pattern generation and stylized image design. The system begins with self-organized patterns generated by reaction diffusion. To extend the style of reaction diffusion, the proposed method involves using a set of modifications of anisotropic diffusion to deform shape and introducing a flow field to guide pattern arrangement. A pattern picker is proposed to facilitate the pattern selection from these modifications. In the post-process step, a new thresholding and color mapping method is introduced to refine the sizes, densities, and colors of patterns. From the experimental results and a user study, several image stylizations, including paper-cut, stylized halftone, and motion illusion, are generated by using our method, demonstrating the feasibility and flexibility of the proposed system.
