Blending mechanism of optimization-based colorization

Optimization colorization methods base on a specific local relation model, it can not afford to describe global characteristics such as direct connection between visual result and user-provided scribbles. Hence a general frame is put forward, and then it is proved that two seemingly different methods are the same case of it, therefore they are equivalent. Next, it is derived that colorization result can be synthesized a weighted sum of chrominance of all kinds of scribbles where coefficients are all nonnegative and normalized and it leads to the conclusion that optimization-based colorization is an exact process of chrominance blending, and moreover, equivalence between individual chrominance blending weight and the pixel's first-reach probability to relevant scribble is determined. The mechanism inspires some improvement to existing algorithm. Experiment results indicate that introduction of blending mechanism benefits visual result to user's scribbles not only in robustness to amount and position but convenience in selecting desirable colors in-time.