Started as Python OpenCV experiment to make a collage of album art. Began to explore methods to automatically determine order. Implemented simple linear classifiers with promising results, training order based upon input from k-means segmentation of the images in LAB color space.
- numpy + matplotlib
Clone into a directory
git clone https://github.com/vinard/AlbumArtPosterMaker.git
Enter directory and make directory for source images
cd AlbumArtPosterMaker mkdir src_images"
Gather album art from somewhere (eg., AlbumArtExchange), and save images into directory for source images. Works best with square images of at least 300x300px.
From top level directory, run:
python3 color_ordering.py ./src_images
Sit back and wait! Collage should be written in top level directory after images' dominant color is assessed
Future Work (TODO)
- Gather human ordered data (Mechanical Turk, 3Blue1Brown sponsor, etc.)
- More robust classifiers
- Read suggested materials from Prof Nayar OH
- Continue updating README to follow good practices template
11/01 Office Hours with Prof Nayar
- Color Indexing by Swain and Ballard
- Will get far
- Easy way to do search
- Constrain dimensions (255->10) and normalize
- Histogram is reduced
- Easier to train on
- Histogram intersection
- Not just sorting, also matching
- 3 images, center is closer to which?
- Mechanical Turk (Amazon)
- Or build on website and send to people
- Scrape all album covers from website
- Set up interface for comparison