I'm working on an algorithm to detect times new roman adjacent fonts lol.
Started off as a joke about the "None So Vile" album cover using the font. Turned into a strong desire for being able to identify times new roman in my spotify library.
It's mostly just an image processing task, with a bit of graph theory. Assisted by the spotify api.
I'm playing around with some different approaches. But the idea is to do template matching, either on the entire image, or using OCR to find bounding boxes. Then use that to align letter masks to the image in the "best way". I.e. good matches, similar font sizes, correct order of letters, similar and reasonable horizontal spacing, etc.
I can't really decide on the best measure for determining the quality of the match, so I'm probably just gonna use several haha.