Research Of License Plate Recognition System Based On Wavelet Transform

With the rapid development in China’s economic make the car ownershipincreased,highway traffic has become a major transport pathway and it becomes theinfrastructure to develop of the country. The more efficient traffic management is neededbecause of the growth of traffic.Intelligent Transportation System(ITS) has become the maindirection of the current traffic management system, which is more efficienct inmanagement,transport and safety.In this paper, modern signal processing techniques-wavelet technology is applied to theIntelligent Transportation System license plate recognition technology, which combined withthe improved BP neural network and template matching for classification of the license platecharacters.The main research contents and results are as follows.1, In image preprocessing phase, the paper uses the gray-scaled transformation, medianfilter and neighborhood filter for image noise reduction, uses histogram equalization and graystretch for image enhancement. In order to highlight the license plate information, this paperuses a new wavelet transform algorithm. It is proved that the enhancement algorithm usingwavelet transform can smooth the histogram and overall the image bright, and the edge hasbeen significantly enhanced.2, In processing of license plate location, this paper uses the license plate locationalgorithm dbN wavelet. By studying the characteristics and nature of dbN wavelet, and by thelarge number of simulation experiments, the db8wavelet on the left and right borders of thelicense plate positioning. Db8wavelet decomposition processing the image usingmathematical morphology and other methods to obtain the license plate of the upper and lower boundary, and extract useful information. The test proved that the db8wavelet caneffectively remove the lights and unrelated to the region, it provide good support for thelicense plate of fine positioning, the final positioning accuracy of the algorithm on the licenseplate up to91.2%.3, The character segmentation stage, firstly make the license plate image binarization andthe correction of the tilt angle. By a large number of experiments, the threshold function basedon the wavelet transform to eliminate the noise of the plate area; and character segmentation,remove the license plate up and down the border, use the vertical projection of a prioriknowledge and the way split of each a character area, and then look for the character ofmorphological connected domains, and cut out the character rectangle of minimum area,merge the mistakenly segmentation Chinese character.4, For character recognition, character of image must be preprocessed, including sizenormalization and refinement process; then study the characteristics of the characters and theslow rate of convergence for BP neural network in the process of Chinese characterrecognition,based on the improvement of the BP neural network to identify; an improvedtemplate matching algorithm be put, template matching combined with character features andextract the edge of the template to identify the numbers and letters. This method allows thecorrect rate and adaptability have been improved, with a good application value.