- The main use of the following two papers "slide window flitering" and "colorization using optimization", for the above code, the implementation of the use of Python methods as well as the use of Laplace operator optimization(strengthen the edge and reduce the effect of the coloring bleeding), the experimental comparison results do have a certain effect of enhancement. So finally, the Laplace operator optimization is adopted.
- colorzationOptimization.py is a reproduction of the original paper on colorization using optimization, which is mainly to solve the system of linear equations in YUV space with the help of the theory of the above paper.
- slideWindowColorization.py is a reproduction of side window filtering + Laplace enhancement to achieve colorization filling.
-
python slideWindowColorization.py --padding 2 --gary_photo_file ./data/original/example1.png --marked_photo_file ./data/marked/example1_marked.png --is_file # choose the single photo to colorizate python slideWindowColorization.py --padding 2 --gray_data_dir ./data/original --marked_data_dir ./data/marked # choose all files of folder to colorizate
-
paramter content padding Side window window radius size, default is 2 gary_photo_file When setting --is_file you need to set the gray scale image name marked_photo_file Marked image name is required when setting --is_file gray_data_dir Default setting is to read gray image folder path marked_data_dir Default setting is to read marked image folder path is_file False by default exp_dir Defaults to . /data/exp is_reshape Defaults to False is_store when --is_store is in parameters list, the data should be stored in exp_dir, and its default setting is false
- When doing RGB to YUV space, it was found that dividing the RGB value by 255 works better than the maximum value of the enhanced RGB when normalizing RGB.
- Links to specific papers as well as references are shown below.