使用者引導影像線段的抽象化

User-Guided Abstraction of Image Strokes

林佑臻(Yu-Chen Lin)

指導教授 ： 李潤容

國立清華大學/電機資訊學院/資訊工程學系/碩士(2013年)

摘要

在非相片質感繪圖的研究中，如何讓使用者可以畫出如畫家般有技巧的線段，能夠兼顧影像特徵與抽象化，一直是很困難的問題。現行的幾種抽象化演算法都需要調整各項參數，產生不同的細節以適應不同的影像，這和畫家創作概念不相同，一般使用者也會覺得不直覺。
在本研究中，提出了一個輔助繪圖系統，可以讓使用者引導進行影像的線段抽象化，讓沒有畫畫經驗的使用者也可以快速畫出有風格的圖像。本研究基於Winnemöller, Kyprianidis和Kang等人所提出延伸型高斯差演算法進行邊緣偵測，進行有粗細的線段向量化構成的線段資料庫。當使用者繪圖的時候，進行資料庫線段比對，即時顯示於畫布之上。如此一來使用者不需要調整任何參數，就可以引導系統速寫作畫。

關鍵字

非相片質感繪圖 ； 筆觸向量化 ； 高斯差 ； 使用者引導

並列摘要

How to let user draw as an artist is always a difficult problem in Non-Photorealistic Rendering field. Several researches on image abstraction had proposed many algorithms, but most of them need to provide different parameters with respect to different images. In this study, we present a system that let the user join the creation of painting just like the artists do in real world.
There are four stages in our system. First two stages are preprocessing and the other two are interactive processing. In the first stage, we apply Different-of-Gaussian (DoG) filter to source image. Then vectorize the image as the associated internal representation. Different DoG results with different parameters were used to build a database for further reference. In the third stage, we calculate matching weight from database when the strokes user drew. And in the final stage, the closest edges from the database were extracted and put into the final canvas.
Our system is very intuitive and easy to use for novice user. User can draw stylish painting with a few strokes in short time.

並列關鍵字

NPR ； Stroke ； Vectorization ； Different of Gaussian ； User-Guided

參考文獻

[Curtis et al. 1997]

[DeCarlo et al. 2002]

DeCarlo, D., & Santella, A. (2002, July). Stylization and abstraction of photographs. In ACM Transactions on Graphics (TOG) (Vol. 21, No. 3, pp. 769-776). ACM.

[DeCarlo et al. 2003]

DeCarlo, D., Finkelstein, A., Rusinkiewicz, S., & Santella, A. (2003, July). Suggestive contours for conveying shape. In ACM Transactions on Graphics (TOG) (Vol. 22, No. 3, pp. 848-855). ACM.