A Novel Method to estimate blur from License plate images of moving vehicles

Abstract

Vehicles throughout the globe are identified by the license plate attached to them. But the automated identification of such plates becomes difficult when the vehicle is moving very fast as it creates a large amount of blur on these images. In fact, such images cannot be even deciphered by humans. The present work proposes a method to deblur such images. The blur can be evaluated as a kernel characterized by angle and length. In the proposed method, the blur kernel is identified based on sparse representation. The sparse representation coefficients can be used to estimate the angle of the kernel because these coefficients will be sparsest when the kernel angle is correct. This can be posed as an optimization problem to find the correct angle of the blur kernel. The length of the blur kernel is then evaluated by the Radon transform.