I have a made up alphabet consisting of around 20 shapes and letters that are similiar to greek and cyrillic letters. They are generated as assets and i have 30 by 30 images for each of them. I want to create a special image processing tool to live translate them. They alway be printed on a black polygon. So i have tried to detect using black polygon with generic opencv2 methods and it was succesful. To scan and detect letters i have tried ORB feature extraction and matching but it was not succesfull. It found features all over the letters due to similar runic shapes. I have tried to train an object detection model with Yolo11 but due to low amount of data ( ı dont have already printed examples, i tried to generate mock images with diffrent angles ) it was costly and underperforming. I dont have enough data to train a HOG for it. Is there a simple pattern matching algortihm for python that considers small tilt pan and roll of camera in real life so can still detect letter? edit -> Here is some examples of letters i have Continue reading...
