Study On Restoration And Recognition Of The Doorplate Number Image Degraded By Linear Motion

This thesis takes the dynamic doorplate recognition as the research object, systematically studies on restoration methods of blurred images degraded by linear motion between a CCD camera and a doorplate, as well as on recognition of doorplate numbers. The main contents can be concluded as follows.1) A novel restoration algorithm for composite-frame motion-blurred images is put forward. The algorithm includes two stages. First, eliminating the edge"staircase effect"based on a block-matching idea; second, removing the motion blur in each sub-field based on the Wiener filter. Experiments validate the proposed algorithm through restoring the blurred images captured by relative linear motion between a CCD camera and a doorplate mounted in a vehicle with different relative speeds and distances.2) On the basis of the introduction to two implementation methods of Markov Chain Monte Carlo (MCMC), i.e., the Metropolis-Hastings algorithm and the Gibbs sampler, the detailed MCMC implementation procedure and experimental results of restoring relative-motion-blurred images are presented.3) The back propagation (BP) neural network is successfully applied to identify printed door-numbers with different fonts.