Skew detection and correction of Indian vehicle license plate using polar Hough Transform research

Abstract

One of the major bottlenecks in an Automatic License Plate Recognition system (ALPR) is the presence of skew and shear in the captured localized license plate due to the position of vehicle with respect to camera. Skew not only spoils the LP character image segmentation accuracy, but makes the segmentation impossible in quite a many occasions. Skew Correction is a pre-process before License Plate Character Segmentation. Proper skew correction after localization will enables near perfect character segmentation and recognition in ALPR process. There are several techniques available to correct the skew of the localized LP plate including Projection Profile, Threshold method, Connected Component analysis, Centroid method, Convex Hull method, Morphological method, Radon Transform method. This paper proposes a new skew correction algorithm using polar Hough Transform. Major problems faced by most of the skew handling algorithms like restriction on the detectable angle range, restrictions on type or size of fonts, dependence on License Plate layout, presence of dirt and dust on the License Plate and high computational cost are minimized due to the concept of luminance threshold introduced in this algorithm, which correctly identifies the pixels whether black or white. The performance of the proposed algorithm has been tested on real Indian License Plate images, which shows superior performance in car license plate segmentation in Indian ALPR context.