An optimized license plate recognition system for complex situations

This paper optimizes traditional license plate recognition system (LPRS) and for each original part with deficiencies, we give our own novel methods to refine them, such as integrating color and edge detection to increase the success rate of locating LPs as well as employing connected component analysis and vertical projection method alternatively to make segmentation more precise and efficient. For character recognition, we apply improved K-Nearest Neighbors algorithm and introduce some novel feature vectors to improve the accuracy of recognition. Our experimental results indicate that the optimized system has a high LP recognition rate with the accuracy of 96.75%.