| Summary: | 碩士 === 國立中興大學 === 資訊科學系所 === 94 === The existing colorization techniques can roughly categorize into automatical, manual, and hybrid approaches. This thesis presents a new algorithm to colorize gray-scale images. We also extend our algorithm to colorize the video clips. Our algorithm belongs to the hybrid ones, requiring no segmentation and texture mapping technique. The algorithm contains four procedures: (1) the image sampling step: producing micro sample points both in the grayscale and reference images. (2) micro-point colorization: searching the best luminance match to colorize the sampled micro-points; (3) macro-area colorization: employing the state of the art colorization methods to draw colors on the entire image; (4) user adjustable re-colorization (UAR): adjusting colors in the inaccurate areas to make the whole image more natural.
We implemented our algorithms and conducted the experiments using several gray-scale images and video clips. The experimental results demonstrate that our algorithm is feasible to colorize the grey-scale image, producing visually plausible results. Also, the visual appearance of the colored video clips is similar to those produced by our counterparts, but our algorithm requires less processing time and less elaborative efforts for the colorization. Our algorithms have four advantages: (1) automatic colorization; (2) rich color selection for the colorization; (3) low colorization error rates; (4) fast processing time.
|
|---|
