Deep Learning-based approach for Indian License Plate Recognition using Optical Character Recognition

The registration number (license number) of vehicles can be automatically read from digital images using a method called Optical Character Recognition (OCR). The enhanced license plate recognition method described in this paper uses a neural network trained on an alphanumeric dataset. The whole system can be categorized under two major modules, namely Plate Character Segmentation, and Plate Character Recognition. In this work, the pre-trained ResNet50 model is fine-tuned. The R
$e$
sNet50 model is trained (fine-tuned) using the custom dataset after the top layer is removed and a new layer is added to the network. A Convolutional Neural Network (CNN) model is also trained on the same dataset and the results of both models are then compared. From the results, it can be inferred that the proposed CNN network with a validation accuracy of 95.60% outperformed the pre-trained ResNet50 network. Lastly, the lower computation time of the proposed network makes it extremely suitable for real-time applications.