In this paper, we present a robust recognition method for license plate numbers that are severely distorted by down-sampling, low light level illumination, and geometric distortion. The proposed method adopts a novel approach, where a set of reference license plate images are intentionally degraded using the same degradation factors estimated from the input distorted images. We first generate a set of reference images using notch filter, band-pass filter, and geometric transformation. We then compare input images with the degraded version of the reference images in the sense of sum of absolute difference (SAD). The proposed method provides an acceptable recognition rate for severely distorted input images that cannot be recognized by human vision.