Re-colorization with random walks

Abstract

Image re-colorization means replacing the original image colors with other

specifies colors. Image segmentation is the technique separating an image into

different regions according to likelihood between adjacent pixels. My final year

project conducted a research on combining random walks image segmentation

and chrominance blending image re-colorization. The objective of this project is

developing stand alone program realizing image re-colorization with

segmentation performed first. The results are expected to be as good as the

existing approaches. The program development was started after my solid

understanding on the background knowledge. It took me around 6 months to

finish the project. I used related papers and available source codes as the key

references. I also adopted GMM in the random walks algorithm.

The program developed gives quite accurate segmentation result, with

natural re-coloring effects. Compared with best existing approach, it shows less

sensitivity to users’ input. However, the speed of the program needs to be

improved by various means.

In the report, the introduction includes overview of the project, the project

purposes, background information and limitations. Then it will discuss on design

consideration as well as implementation details. After that certain test results will

be presented to comment on program’s robustness. Last but not least, the

conclusion wraps up the whole paper with meaningful recommendations on

future works.