New Platform for Automatic Iranian License Plate Detection and Recognition using Deep Learning Techniques

In our modern world where the population of the cars and vehicles are growing fast, controllinsssg is an important issue. There are lots of methods which can be used for identification of criminal drivers and penalize them, but the least cost and fastest method will automatically recognize by capturing pictures from vehicle path such as freeways and streets. To date, many researchers around the world, are proposing and improving the algorithms which can show notable, fast and accurate performance. These efforts make the license plate (LP) detection very interesting in the image processing field. In Iran, the rules for LP numbering is different from other countries. The plates in Iran consist of two digits in left, an alphabetic character in middle (which shows the purpose of car such as transportation, military, personal, and so on),three digits on the right side, and finally two digits which show the province of the car. There are many proposed algorithms which have segmented and recognized Iranian License plate, but they have some problems during the segmentation and recognition steps. In this article, we try to solve previous problems with our new algorithm. We Apply some preprocessing methods and a combination of morphological and spatial filters in plate segmentation section and a well performed deep neural network using Convolutional Neural Network (CNN) model to recognize all digits (0-9) and characters (16 Alphabetic). Finally, to achieve a user-friendly interface, a Graphical Unit Interface (GUI) isdesigned with MATLAB. The performance of the purposed model is tested with two datasets which contain the images under various conditions, such as poor capturing quality, image perspective, rotation distortion, sunny day and night. Our new algorithm shows high performance in both Iranian plate segmentation and recognition with 98.77% accuracy on test plates

Keywords:

Iranian license plate detection , License plate recognition , Segmentation , Deep learning , Tensorflow , Convolutional Neural Network , Graphical unit interface