Research On Vehicle License Plate Recognition Technology Under Complex Background

Vehicle License Plate Recognition System(VLPRS),developed in recent years, is an intelligent traffic management system based on image processing and pattern recognition, which plays the most important part in Intelligent Transport System(ITS).Generally, the key techniques of VLPRS include image acquisition, image preprocessing, Vehicle License Plate Detection(VLPD),character segmentation and character recognition. To meet the growing practical necessity, related key technologies of VLPRS still need be further studied. After we'studied in depth on the key technologies of the existing VLPRS, we presents a practical, advanced and integral vehicle license plate recognition processing and algorithms, then we validate them under VC6.0 software environment.In the aspect of VLPD, we use an algorithm based on the morphological operations to seek approximately location:firstly, form the license plate region, which is rich in texture characteristics, to a whole connected region by morphology means; and then use the method of connected region labeling and selection to locate the license plate roughly; finally, remove false license plates by the vertical projection of the binary image of candidate plate regions.In the aspect of character segmentation, we use the method of vertical projection of the binary image of license plate combined with prior knowledge to divide characters in the license plate:use the method of vertical projection for rough segmentation and then deal with character adhesion, fracture, disturbance of license plate frame by making use of prior knowledge.In the aspect of character recognition, according to the complex stoke but low resolution characteristic of Chinese character, we use an Uniform Local Binary Pattern(ULBP) operator for feature vectors extraction, and then identify the Chinese characters using the method of histogram matching; on the part of non-Chinese characters, English letters, Arabic numbers and letters-numbers-mixed neural network are established; finally, structure characteristics was used to differentiate the similar character. For the above process and algorithm, experiment for the actual image was done and preferable result was achieved.