License plate recognition with feature salience and neural network

Abstract

Character recognition algorithm is considered as a core component of License Plate Recognition (LPR) systems. Numerous methods for License Plate (LP) recognition have been developed in recent years. However, most of them are not advanced enough to recognize in complex background and still demand improvement. This paper introduces a novel system for LPR by analyzing vehicle images. Accurate segmentation of license plate and character extraction from the plate is accomplished. In the plate segmentation module, Hough transform is put forwarded to identify plate edges using line segments. Radon transform adjusts the skew between LP and the viewer, thereby improve the recognition result. Four features are extracted from the LP image, and best features are selected using feature-salience theory. Histogram projection is performed horizontally and vertically to isolate individual characters in the LP. Finally, Back Propagation Neural Network (BPNN) is used to identify the characters present in the LP. From experimental results, it is evident that the proposed system can recognize LP more efficiently and establish a good background for future advancements in LPR.