Python Image Processing Cookbook

There is more

For feature-based similarity comparison between two images, you could use the proportion (percentage) of matches as the ranking criterion too (for example, have the percentage of a match in the keypoints as key in the matched_images dictionary). Also, for scalability and reusability, you should serialize the descriptors for all of the search images you have in your search directory (for example, use pickle) and load/deserialize all of the descriptors when you want to match with a query image descriptor—it is also left for you as an exercise. You can use SURF, KAZE, or any other feature for matching and any distance/similarity metric between the descriptors. For production, use an algorithm for fast computation of the similarities ...