Practical Homography-based perspective correction method for License Plate Recognition

Abstract

Automatic License Plate Recognition (ALPR) can avoid faults of manual license plate recognition, like pressing keys wrongly or too slowly. But, there are inevitably some vertical and horizontal perspective distortions between the license plates and ALPR's cameras, degrading the accuracy and reliability of ALPR significantly. This paper proposes a Homography-based perspective correction method for ALPR. Especially, in order to overcome three variation issues residing in ALPR systems and applications frequently, this paper further proposes three practical auxiliary methods: 1) YCbCr color space differentiation to overcome the background color variation (e.g., white, green, or red) on license plates, 2) sub-regional histogram equalization to overcome the frame contrast variation between the license plate surrounding and the vehicle body (e.g., silver and white-like), 3) diagonal- and Houghlines-scanning four-corner localization to overcome the frame shape variation of license plates (occluded by stains or reflections). Experimental results show that the license plate perspective correction rate of the proposed method for automotive and motorcycle license plate database are 98% and 94%, respectively. And, after corrected by the proposed method, license plate recognition rate for automotive and motorcycle license plate database are 97% and 89%, respectively. The proposed perspective correction method for ALPR is more useful and reliable at solving real-world perspective distortion issues than conventional ones.