Robust Feature Extraction Technique for License Plate Characters Recognition

This paper presents two novel geometrical feature extraction techniques for license plate (LP) characters recognition, written in English alphabets and numerals. A robust character recognition technique has to recognize different format, broken, and distorted characters by extracting invariant and discriminative character features. In this paper, we have extracted two types of geometrical features and a statistical crossing count feature with the help of horizontal, vertical, right diagonal, and left diagonal scan lines and centroid of the character to be recognized. For character recognition, we have used edit distance metric. The proposed technique can be enhanced to recognize any language scripts and is tested for Indian scripts. The proposed LP characters recognition technique is evaluated using 741 images of media-lab benchmark database, 49 images of Israeli vehicle LPs, and 110 images of proprietary Indian vehicle LPs having different environmental conditions and plate variations. The proposed technique achieved success rate of 98.8% for LP characters recognition which outperforms many LP character recognition techniques in the literature.