以真實影像為基礎之水墨畫風格模擬

Abstract

本研究建構出一個自動將使用者輸入的影像轉換成水墨畫繪畫效果的系統，利用以下六個步驟，輸出具有出自畫家之手的水墨畫風格之影像： 1. 利用高斯金字塔將輸入影像分成八種影像，再透過特徵取樣取出四十二張特徵圖，再將其組成影像顯著圖 2. 利用引導式濾波器，使輸入影像模糊化，去除細節，並保留邊緣 3. 將模糊化影像做水墨渲染效果 4. 將水墨渲染效果之影像，轉換成灰階圖並去顏色化，使其黑色變得更黑，白色變得更白 5. 利用邊緣偵測模型找出輸入影像之邊緣輪廓，用作筆法勾勒 6. 將去顏色化之影像與邊緣偵測圖合成，即為結果圖 透過本系統處理後的影像，即擁有水墨畫風格且如出自人手之畫作風格影像。

This thesis constructs an image-based Chinese ink painting simulation system. Via six steps, the system will output an image that has Chinese ink painting style.

The six steps include: (1) Nine spatial scales are created by dyadic Gaussian pyramids, then compute forty-two feature maps and combine it to a salience map. (2) Using Guided Image Filtering for blurring image. (3) Using diffusion algorithm to simulate ink diffusion. (4) Transfer color image to gray image and black-white enhance. (5) The edge-detection algorithm will figure out the edge of image that is used to draw the outline of calligraphy. (6) Combine the decolorizing image and edge-detection image. And it is the result. The image which is processed by this system becomes the ink painting style image that seems like drawing by a real painter.