Optical Character Recognition (OCR) has been investigated widely to recognize characters in images for various applications including license plate recognition. Several limitations and distortions are available in images such as noise, blurring, and closed characters (alphabet and numbers) which makes the task of recognition more complex. This paper addresses the closed characters and blurring problem utilizing three pre-trained deep learning OCR models including Pytesseract, EasyOCR and KerasOCR. We evaluated and compared these methods using a dataset that contains Malaysian license plates. The results show that KerasOCR was able to outperform other methods in terms of recognition accuracy. KerasOCR was able to recognize 107 images out of 264 images compared to only 87 images in EasyOCR and 97 images in Pytesseract.
An Evaluation of Various Pre-trained Optical Character Recognition Models for Complex License Plates
Advances in Engineering res
Proceedings of the Multimedia University Engineering ; 2022 ; Cyberjaya and Melaka, Malaysia July 25, 2022 - July 27, 2022
Proceedings of the Multimedia University Engineering Conference (MECON 2022) ; Kapitel : 4 ; 21-27
23.12.2022
7 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Optical recognition of motor vehicle license plates
Tema Archiv | 1995
|British Library Conference Proceedings | 1999
|