Automatic Colorization based on Image Segmentation

• Main Authors: Shang-wei Chen, 陳尚緯
• Other Authors: Wei-chang Du
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/56620876909355748969

碩士 === 義守大學 === 資訊工程學系碩士班 === 97 === The main objective of this thesis is to colorize grayscale images automatically by reference to the color information in another color image. Unlike using the traditional pixel-by-pixel matching algorithm, the mean-shift image segmentation is firstly adopted in color image and grayscale image respectively to separate into semantic regions as a basic unit in colorization. Since all pixels inside one region have similar color and texture features, the colorization can achieve the spatial coherence. In order to determine the referring source of grayscale image, use the clustering method in each region to find the main multi-luminances, and match these luminances by the one-way Hausdorff distance to increase the correctness of region matching. Finally, according to the texture feature in a local region, give every pixel in grayscale image an appropriate color to finish the coloring goal. This proposed method is a fully automatic coloring process, including image segmentation, region matching, matching colorization, and a series of actions. Such an approach not only can reduce human intervention and computational time, but also can display a natural colorization fit as human visual experience.