Low-quality License Plate Character Recognition Based on CNN

In this paper, we analyze the importance and the difficulties to recognize car plate characters from low-quality monitoring videos. A solution is proposed to achieve the character recognition by classifying character images based on convolutional neural network. We analyze the degradation of car plate character in practical surveillance videos, and then model and generated simulated dataset. We also train simulated datasets in a convolutional neural network, and applied the trained network to recognize license plate characters in the actual surveillance videos. The experimental results show that this method works well to recognize these car plate characters. The proposed method can help identify some of the license plate characters that human eye can´t identify.