Approach to License Plate Recognition Based on Hidden Markov Model

• Main Authors: Shuo-Chen Jian, 簡碩辰
• Other Authors: Chun-Liang Tung
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/3h8z4g

碩士 === 國立勤益科技大學 === 資訊管理系 === 106 === At present, the license plate recognition technology has been applied to life, such as traffic violation, parking management and vehicle tracking. Therefore, many methods have been proposed for license plate recognition. License plate recognition generally requires determining the way in which character images are described and using appropriate model learning and identification first. Because the license plate characters belong to time sequence images, this paper is expected to propose an HMM-LPR (Hidden Markov Model- License Plate Recognition) method based on the Hidden Markov Model which is often used in time sequence samples, such as speech recognition. The recognition rate of the HMM-LPR is tested through defining character features, learning characters and recognition characters. This experiment uses images which taken by road cameras as samples for training and testing to the detection model, and focuses on adjusting features of image to improve the overall accuracy of the HMM-LPR. The experiment result shows that the image feature size is not suitable for license plate recognition. The main reason is a considerable number of character classes have similar feature representations when character images are transformed into feature sequences, hence the HMM-LPR shows a poor recognition rate. But Hidden Markov shows fault tolerance ability for the same category even it has varied feature sequence. Therefore, HMM-LPR needs to redesign the suitable way of feature conversion method that less information losing but clear distinction between descriptions of different categories for license plate recognition.