In a car license plate recognition system, effective and robust image expansion methods will improve its performance and bring a lower error rate. Two MAP-based super-resolution image reconstruction approaches for single image with a prior image model described as Huber Markov random field are discussed and applied to such a system in this paper. A new spatial smoothness measurement based on a flexible convolution kernel is proposed. Parameters in these approaches are discussed. Improved definition of images and increased recognition rate is also shown through computer simulations.