This repository contains the code for our course project for CS663 - Fundamentals of Digital Image Processing at IIT Bombay. We implemented the following paper/report: http://cs229.stanford.edu/proj2013/KabirzadehSousaBlaes-AutomaticColorizationOfGrayscaleImages.pdf. The published code was written in Python using OpenCV, scikit-learn and pygco. We were unable to execute it since some of the functions were outdated. We have coded our implementation in MATLAB.
- Utkarsh Gupta (@Ug48)
- Yash Shah (@ys1998)
See the report.
For details see the report.
In order to incorporate kNN search into our algorithm, we had to make the following changes - (1) remove quantization of color space, (2) replace SVM classifiers with a kNN search, (3) remove position-based features from the feature vectors and (4) replace the graph-cut labeling algorithm with a weighted average in a-b space. We were able to get significantly better results. Some of them are shown below:
(1) Same image used as reference and test image
(2) Different images used as reference and test image
(3) Multiple reference images
