Parametric Kernel Estimation for License Plate Image Deblurring

Author(s): M. Jagadeeshwari, K. Sridhar

Abstract: License plate plays a key role in identifying the over-speed vehicle or the ones involved in hit-and-run accidents. Even though the surveillance camera captures the snapshot of trouble-maker vehicle, the motion of vehicle during the exposure time would cause the blur of snapshot image with low resolution and can be unrecognizable by human. In this paper, we have proposed a novel approach on the identification of linear uniform motion blur kernel using sparse representation and that can be characterized by two parameters: angle and length. By analyzing the sparse representation coefficients of the captured image, Coarse-to-fine framework can be done for determining the angle of the kernel corresponds to the genuine motion angle. Then, we estimate the length of the motion kernel with Fourier transform followed by Radon transform. Our proposed scheme can handle large motion blur even when the license plate is unrecognizable by human.