Vehicle License Plate Detection Based On Image Sementation

Vehicle License Plate Recognition (VLPR) is the core part of IntelligentTransportation System (ITS) and vehicle license plate (VLP) location is the basis andkey technology of VLPR. Today most VLP location methods are based on image orvideo processing. Due to the environment and lightning, the VLP images may containcomplex background, different kinds of vehicles, various color and light intensity,which severely affect the accuracy of VLP location. How to reliably separate the VLPregion from the complex background and improve the VLP location accuracy is anurgent task.The low rank representation(LRR) based subspace clustering provides a veryefficient method for feature based image segmentation. The method realizes imagesegmentation by clustering the high dimensional feature vectors of an image intoseveral subspaces. Using this method we can separate the VLP region from thecomplex background and thus reduce the interference of the background in VLPdetection and location.In this thesis, we first discuss the characteristics of VLP in our country and thedifficulties of VLP location. According to the characteristics of VLP and thebackground, we analyze the color histogram in Hue-Saturation-Value (HSV) space andpresent an adaptive local binary pattern detection operator, which can provide morereliable texture feature description of VLP. The color histogram in HSV space and theimproved LBP are combined to form the feature vector of vehicle images. We then usethe LRR based subspace clustering method to segment the image features and obtainthe result of vehicle image segmentation. Finally we apply the jump function on thesegmented image to locate the VLP. Experiments on images containing single VLPshow that, this method can effectively reduce the influence of the complex backgroundand improve the VLP location accuracy.