This framework takes a grayscale image and attempts to colorize it. Written in MATLAB.
Given that you have an image dataset (e.g. Caltech 101) in the /dataset directory where this file is executed, colorizeImage.m can run out of the box with the following commands:
ann_class_compile;
colorizeImage(imagePath, train, testMode, numExemplars, levels);
where imagePath is a filepath to the grayscale image, train is a boolean where true will train an entirely new classifier on the dataset, testMode is a boolean where true means the input image is already in color, numExemplars is the number of exemplars to use, and levels is the number of levels to use during multi-resolution synthesis in image analogies.
The following would be an example command to colorize GrayCity.jpg using the existing classifier, 2 exemplars, and 3 levels in image analogies:
colorizeImage('images/GrayCity.jpg', 0, 0, 2, 3);
Read the technical write-up and discussion of results here.