Automatic Low-Quality License Plate Recognition (ALQ-LPR) for Smart Motorcycle Parking

This study compares three image preprocessing techniques to improve EasyOCR motorbike parking license plate detection. EasyOCR is a deep learning-based Optical Character Recognition (OCR) system that can recognize text in images, including low-quality or distorted vehicle license plates. Techniques This study compares three picture preparation methods to improve EasyOCR motorbike license plate detection. Otsu's Thresholding, which automatically selects the best threshold value to provide clearer binary images and aid OCR text recognition, is initially used to analyze low-quality license plate photos. Second, global thresholding uses a consistent threshold value throughout a picture to convert it to binary. Images with little contrast or noise perform poorly using this approach. Third, adaptive thresholding adjusts the threshold based on local pixels, which should perform better for uneven lighting photographs. The study's findings demonstrate that Otsu's thresholding, which is successful for low-quality photos despite occasional detection errors, performed the best with 100% precision, 87.5% recall, and 80.36% accuracy. With 50% precision, 14.29% recall, and 35.72% accuracy, global thresholding performs worse and has trouble identifying low-quality license plates. Although adaptive thresholding struggles with distorted pictures, it exhibits 100% precision but relatively low recall (12.5% and 28.58% accuracy). All things considered; Otsu's thresholding is advised for OCR systems on low-quality motorbike license plates. The next stage of the project is to test the pre-processing techniques of bilateral filtering and Contrast Limited Adaptive Histogram Equalization (CLAHE), which may be used to reduce noise and make the text on the hazier license plates more readable.