License plate image restoration and recognition method based on context encoder

This paper proposes a contextual encoder-based license plate image restoration and fast character recognition method, which aims to solve the problem of license plate recognition in complex scenes such as blurred and occluded license plate images. The algorithm is mainly composed of an image restoration module and a character fast recognition module. The image restoration module adopts the context encoder structure, and trains the repair model through the encoder-decoder architecture combined with L2 loss and adversarial loss, which can effectively repair the missing parts in the license plate image and improve the image quality. The character recognition and classification module adopts the improved Yolov5S algorithm to optimize the scale characteristics of the model, reconstruct the multi-scale detection grid of the original model into a linear distribution structure of 1×7, and realize the positioning and classification of license plate characters through a single forward propagation, which significantly improves the detection speed and accuracy. Experimental results show that the proposed context-based encoder performs well in the image restoration task, and the restored image is close to the real scene in terms of texture detail and structural rationality. Compared with Yolov5S, the character fast recognition method proposed in this paper has an improvement of 4.3% in [email protected] and an accuracy rate of 93%, which meets the needs of real-time license plate recognition. The algorithm in this paper has significant advantages in license plate recognition tasks in complex scenarios, and is suitable for practical application scenarios.