License plate (LP) segmentation can easily be affected by rim, noise and LP rotation, which causes low segmentation accuracy. We presented a goodrobustness segmentation algorithm for the issue. The algorithm performed brightness classification and grayscale enhancement in the preprocessing stage before tilt correction and localization of the upper and lower boundaries. An improved dichotomy algorithm was then applied to segment LP characters. We implemented precise border localization for characters 1, 2, 6 and 7. Experiment shows that the algorithm has low computational burden and can effectively overcome such negative influences as low contrast, blurness, character adhesion and LP tilt.