Priori embedding for character segmentation in license plate recognition(LPR) is difficult

and the segmentation results greatly depended on the precision of license localization. Thus in this paper

we present a novel integration method of the prior knowledge and its corresponding character segmentation algorithm. Given one category

the segmentation procedure can be transformed to a set of Markov transitions where the state is defined by the characters permutation. With the combination of connected-component-based and vertical-projection-based algorithms

the optimized segmentation result and its confidence level can be obtained. One advantage of our method is that only coarse localization is enough for the segmentation algorithm. The proposed method is uniform for different license categories

which reduces the coding complexity. The performance improvement of the proposed approach is illustrated by the LPR systems applied practically in China and Malaysia

both contains multiple categories of license plates.