An efficient robust method for accurate and real-time vehicle plate recognition

Accuracy and real-timeliness are the top concerns in vehicle plate recognition. Several factors put restrictions on plate recognition system, including illumination, vehicle high speed, camera angle, and bad weather condition. Damaged and pale plates also lead to incorrect recognition in the present approaches. In this regard, this paper proposes an efficient robust method for vehicle plate recognition, which consists of four steps: (i) vehicle detection, (ii) plate detection, (iii) character segmentation, and (iv) character recognition. In the first step, the vehicle image is detected using background emission. Plates are localized by means of character recognition and pattern matching approaches in the second step, where the contours are recognized and extracted using connected component analysis, and then, low-density areas are emitted using density criterion and vehicle plate is extracted. In the third step, statistical feature, filtering methods, and morphology operators are employed for segmentation and extraction of plate characters. After plate segmentation, statistical and global features and local pattern are extracted from each segment image for segment classification in the final step, where features are ranked using F-Score, and then, classification of each section to one of 37 classes is performed using random forest. The proposed method is evaluated using several databases in both left to right and right to left languages; English for the former and Persian for the latter. In the first part of the evaluation, the proposed approach is evaluated in terms of robustness and recognition speed. The proposed method has the accuracy of 99.2% for plate recognition, 100% for plate segmentation, and 98.41% for character recognition. In this part, the dataset of Iranian plates is collected by the authors of this paper. However, character recognition rate is 100% in other Persian databases. Moreover, the experimental evaluations witness that the proposed method can process at least 8 frames per second, that means it is fast enough to be adopted for real-time applications. In the second phase, the proposed method is evaluated on an English plate dataset. In this dataset, the proposed method shows an accuracy of 100% for plate detection and 97.5% for character recognition. The experimental results show that the proposed method outperforms methods proposed in recent years in terms of time and accuracy that is also independent of plate language.