| Summary: | 博士 === 國立中興大學 === 資訊科學系所 === 94 === This dissertation presents three color transfer algorithms for images. The first algorithm is an automatic, swatch-based, color transfer (ACT) algorithm for two still images. It modifies colors in the source image by borrowing the color characteristics from the target image. The algorithm consists of three processes: a swatch generation process, a swatch matching and color transfer process, and a swatch boundary transition process. This swatch-based algorithm proceeds on color transfer with no user intervention, and produces visually plausible resultant images. Experimental results demonstrate that the ACT is superior to Reinhard et al.’s original user-intervention color transfer algorithm.
The second algorithm presents a novel automatic color transfer approach for image sequence (ISCT). This ISCT algorithm renders an image sequence with color characteristics borrowed from three user-given target images. The algorithm completes the color transfer task with three processes: a forward color space conversion process, an image sequence process, and an image sequence animation process. Given a single input image (I1) and three target images (T1, T2, T3) as inputs, the algorithm produces an image sequence {Si} with color mood variations. The ISCT algorithm is fast. It achieves the goal of rendering an image sequence in several seconds. In addition, the user interface developed provides much freedom to visualize the rendered image sequence. Given only three target images, the ISCT algorithm demonstrates its feasibility to produce an image sequence with visually plausible effects. This algorithm is automatic, effective, and expeditious, and is appropriate for many applications.
Finally, we recommend a generalized color transfer algorithm for image sequences (GISCT). There are two major differences between the GISCT algorithm and the ISCT algorithm. The first major difference is that the GISCT algorithm proposes a new color transfer approach (NCT) to eliminate the appearance of over-transformation, which occurs when the source and target image are not compatible. The second difference is that we present a generalized color variation curve (GCVC) in the GISCT algorithm. Specifically, a B-spline curve is automatically generated to interpolate color statistics, which provides more flexible control over in-between images. Experimental results show that the GISCT algorithm generates results in several seconds. It renders an image sequence with versatile color variations, producing more visually plausible appearance than those generated by the ISCT algorithm.
In conclusion, the ACT algorithm produces results that are superior to its closest competitor. The novel ISCT and the GISCT algorithms generate an image sequence with color mood variation. These algorithms contribute significantly to the topic of color transfer in the computer graphics community.
|
|---|
