Russian License Plate Segmentation Based On Dynamic Time Warping

We propose a fast plate segmentation algorithm for automatic license plate recognition systems which is stable to plate rectangle localization inaccuracies and image brightness distortions. The algorithm uses a priori information about the geometry of standard plate types and additionally makes adjustments to symbols positions through localization errors estimation and correction. We introduce a plate localization error model and compute its optimal parameters using dynamic programming. We also suggest a modification of the algorithm for simultaneous segmentation and optimal type selection (from a known set of types). Experimental results are presented which show the efficiency of this approach.