Performance Evaluation of Tesseract and EasyOCR to Recognize Bangla License Plate for Diverse Quality Image

With several applications in traffic management, safety and vehicle movement tracking, autonomous systems for recognizing license plates are pioneering developments in computer vision. The accuracy of automated license plate recognition (ALPR) systems has significantly increased due to a revolution in computer vision brought about by the availability of enormous datasets and current advancements in deep learning techniques. Recognizing license plates from low-quality photos presents several challenges such as poor quality, a variety of plate typefaces and changes in size and orientation all contribute to a reduction in character recognition accuracy. In this paper, we have utilized the image enhancement technique to reduce the negative influences of these challenges to recognize Bangla license plates accurately for diverse quality images such as blurred and low-resolution. In our research, the EasyOCR and Tesseract OCR frameworks, along with the string-matching approach are applied to recognize license plate characters. We have evaluated the performance of EasyOCR and Tesseract OCR and observed that the performance of EasyOCR is higher compared with Tesseract OCR. For blur images, EasyOCR has given 76.05% character recognition accuracy and it has taken 0.128 seconds to recognize text from a license plate whereas Tesseract OCR has given 47.90% character recognition accuracy and it has taken 0.175 seconds to recognize text from the license plate. For low-resolution images, EasyOCR has given 93.76% character recognition accuracy and it has taken 0.114 seconds to recognize text from a license plate whereas Tesseract OCR has given 50.41% character recognition accuracy and it has taken 0.175 seconds to recognize text from the license plate.