| Summary: | 碩士 === 國立中興大學 === 資訊科學系所 === 94 === Existing colorization techniques produce visually appealing results for natural images such as trees and portraits, but they are not suitable for Chinese ink paintings. This is because Chinese ink paintings have individual features, such as voids, subtle blank parts, and small interstices among strokes. As a result, colorization often produces images with the so-called over-diffusion effect, where colors are diffused in exaggeration to another distinct region. This paper presents a novel colorization algorithm for Chinese ink paintings. Our algorithm can colorize the grey drawings of Chinese ink paintings and remain their important feature appearance. In particular, we employ the concept of the image mask to produce decent colorization results. Our algorithm contains three parts, the fuzzy segmentation part, the colorization part, and the over-diffusion removal part. In the first part, we employ the latest colorization technique introduced by Yatziv to colorize the paintings, and produce an image with the over-diffusion effect. In the fuzzy segmentation part, we first adjust the brightness of the paintings. We then use a fuzzy segmentation technique to properly separate the background from the foreground, generating the background mask. In the final part, we filter out the superfluous parts of the over-diffusion effect using the background mask, ensuring the individual features intact. Experimental results demonstrate the feasibility of our algorithm for colorizing the Chinese ink paintings. Our algorithm has four advantages, including easy colorization, visually pleasing, fast processing, and remaining the important characteristics of Chinese ink paintings.
|
|---|
