Project for Image Processing Course. The goal of the project is to recognize polish license plate from an image.

• the angle between horizontal surface and license plate is +- 45 degrees
• longer edge of license plate is greater than 1/3 of image's width
• license plate has 7 characters
• images can have diffrent resolutions

• written in Python 3.7 using OpenCV
• there's an option to use other libraries (like scikit-image) but you cannot use external OCR modules or trained models that can read characters
• maximum time for each image processing is 2 seconds

• Bounding box accuracy of license plate: 96,42% (27/28)
• Bounding box accuracy of characters in license plate: 91,32% (179/196)
• Total score: 86,42% (Every correct read character on license plate -> 1 point. Correct read of whole license plate equals to 1 point for each character + 3 additional points.)

• Total score: 63.70% (Every correct read character on license plate -> 1 point. Correct read of whole license plate equals to 1 point for each character + 3 additional points.)
• execution time per image: 0.11s

• Create virtual environment using python3 -m venv venv
• Install requirements from requirements.txt
• Run script with python main.py <path_to_folder_images> <path_to_json_results_file>

• convert to gray scale
• resize the image for faster processing
• blur image and find image edges.

• find contours on the image
• create bounding boxes from contours
• skip bounding boxes that doesn't match license plate height to width ratio
• get corners of potential license plate

• construct destination points based on vertices
• get perspective transform matrix
• warp the perspective getting bird's eye view in a process

• find contours on the image
• get bounding box of each contour
• discard all small bounding boxes and bounding boxes that don't match character width to height ratio
• discard contours in contours

• sort potential characters ROIs from left to right
• compare ROI of each potential character with reference characters and get closest matching
• if there's more ROIs than maximum license plate length, delete character with weakest matching

• if there is no maximum number of characters found, fill empty spaces with "?". "?" are filled in proper space based on distance between each character

• compute distance between ROIs to determine number of characters in first part of license plate

• change forbidden characters to theirs closest counterpart. For example 2 -> Z in first part of license plate