Reza serajeh - Image Processing and Machine Vision Group Amnpardaz Software Company Tehran, Iran

License plate recognition (LPR) systems are one of the main parts of traffic surveillance and have three main steps: license plate detection, character segmentation and character recognition. To evaluate and compare purposed methods in these three steps, a large and universal database needs to be collected as a benchmark for each kind of license plate format, language and font. This article introduces such database for Iranian car license plates. The database is collected from color images in common cameras quality. The database contains three main parts. Firstly, it provides 2×12275 car license plate images for 1008 different categories where there exist two images for each car license plate: the extracted car license plate and the main image containing car license plate and its background. In detail each category introduces a car plate number containing 2×1 to 2×132 car license plate image samples. Then these car license plates are individually segmented and 96013 characters are collected from them. Finally all the characters are labeled in 26 clusters and some low quality samples are eliminated to represent training and test images for learning and testing classifiers. The experimental results highlight the challenges on the database for LPR systems

License plate recognition (LPR), license plate detection, license plate segmentation, Iranian and Persian license plate database, intelligent transportation systems