License plate character segmentation algorithm based on variable-length template matching

Abstract

This paper proposes a variable-length template matching method for license plate (LP) character segmentation, according to the characteristics of horizontal and vertical projections. Firstly, it corrects skew and shear distortion with LP boundary and character arrangement rule to get frontal-parallel view of LP regions. Secondly, it projects horizontal gradient of LP along horizontal direction, and uses a set of variable-length single square wave templates to match the horizontal projection of LP's horizontal gradient, thus, to get the top and bottom boundaries of all LP characters. Then, according to the ratio between character with and character space, it designs another set of variable-length templates with seven square waves to match the vertical projections of LP's vertical gradient, and solves the left and right boundary of all LP characters simultaneously. This method can deal with non-uniform illumination, scale change, perspective distortion, image deterioration, background interference and so on, and it can achieve accurate boundaries of LP characters effectively.