Study Of Vital Technologies Of Image Colorization And Color Transfer

Colorization is the process of adding colors to grey-scale images film orvideo. The state-of-the-art colorization methods can mainly be categorizedinto two groups. The first group is example-based colorization. The secondgroup is scribble-based colorization. Gray-scale film and video program willhave a better visual effect after colorization. Therefore, colorization has agreat social and commercial value. Color transfer is the process of carryingover image colors from one image to another. Color transfer has been widelyapplied in a lot of multimedia application fields, such as post-processing offilm and TV program, color correction and color enhancement. Althoughcolorization and color transfer have many applications, error propagation ofcolor、corrupted or noised gray scale image colorization and color transfer vialocal texture mapping always are big challenges for colorization and colortransfer.Firstly, error propagation of color on scribble-based colorization isinvestigated. During scribble-based colorization, chrominance ofnon-scribbled pixels are often dermined by the spatially closest scribbles inthe image. Error propagation of color will occur when non-scribbles and itsclosest scribbles belong to different color regions in the image. In order tosolve this problem, we propose a novel scribble-based colorization algorithmunder the assumption of correlation between luminance and chrominance.The chrominance of non-scribbled pixels are computed by Bayesian MAPprobability. In the Bayesian framework, MRF is used as prior and likelihoodis obtained by Non-Local-Mean coefficients between non-scribbles and theirneighbors. Experiments show that the performance of our colorizationalgorithm is to the existing methods.Secondly, colorization tasks for corrupted or noised gray scale image areinvestigated. Traditional colorization methods can not restore the missing information of corrupted image. We proposed a new united completion andcolorization algorithm based on bilateral filtering technology. Imageprocessing priority is obtained based on distance priority. That means thepixel is earlier processed when it is closer to scribbles. Scribble chrominanceis reasonably propagated according to geometric and radiometric weight ofbilateral filtering. It is noted that adaptive weight selection scheme is usefulto reduce redundant computation. Experiments show that our algortithm isbetter than tradional colorization method, and achieve superior performancethan combinatory image completion and colorization method in the terms ofPSNR and CPU time.Finally, color transfer via local texture mapping is investigated.Tradional color transfer methods are based on statistical features of inputimage. It is a kind of global color transfer method without considering localtexture features. Thus, we proposed to conduct color transfer via local texturemapping. Our algorithm mainly contains feature points extraction and featurepoints mapping. Feature points extraction is obtained based on a pyramiditerative update algorithm from high resolution to low resolution. Featurepoints mapping is obtained by LBP local texture mapping. The initial featurepoints are selected by SIFT algorithm. Pyramid iterative update procedure isdependent on the image residue between original image and colorized image.Experiments show that our algorithm achieves better subjective visual effectand color rationality than the existing methods.