Recognition of severely distorted license plate numbers: Part 2 applications

This paper evaluates the performance of recognizing severely distorted license plate numbers in the sense of resolution, illumination, and geometric transformations. The proposed recognition method first estimates degradation parameters from an input image. Next, a set of reference images are transformed using the estimated degradation parameters. The recognized number is then obtained by comparing the input image with the transformed version of reference numbers. Experimental results demonstrate this work enables to recognize numbers acquired in a critical condition.