License Plate Characters Recognition Using Color, Scale, and Rotation Independent Features

Abstract

This paper proposes a color, scale, and rotation independent approach for recognizing License Plate (LP) characters written in English. We extract features based on geometrical properties of an LP character, like the number of perimeters, perimeter angles, different characteristics points on the perimeter, etc. These features are next used to generate a characteristic encoding of the character. The generated encoding based on color, scale, and rotation independent geometrical properties is then used to identify an LP character independent of its color, scale, and rotational angle. We have evaluated our proposed approach using images from the publicly available Media Lab LP dataset, and a recognition rate of 98.29% was obtained. The processing time for an input image of size 200 × 100 pixel was found to be 0.3 ms. The recognition rate and low processing time compare favorably with results published in the literature by other LP characters recognition techniques. Further, unlike other published techniques, our approach places no restriction on the size, color, or the number of characters in LP, nor is our method restricted to any particular LP design or the LP of any specific country or region.