Rotation correction for license plate recognition

Abstract

The license plate recognition (LPR) is always in operation with the needs of both quantity-and-quality-based approaches. The entire procedure of license plate recognition consists of six steps: image acquisition, image processing, plate locating, character segmentation, character recognition, result output. In real instances of license plate recognition, when a road is uneven with bends, a vehicle will be running with shaky. Consequently, the plate is also unstable and tilted with rotations. Generally, because a surveillance camera is fixed to capture a high-quality image, the plate is hard to be located and recognized. Due to these existing problems, the contributions of this paper are (1) rotation correction for license plate using Hough transform; (2) GNN (Genetic neural network)-based license plate recognition. The novelty of this paper is to recognize a license plate through rotation correction, especially when the border of this plate is not available. Therefore, we detect the straight lines passing through the top and bottom edges of the plate characters and treat them as the border lines. Our experimental results show the proposed method is robust and reliable in LPR correction and recognition.