A Study On License Plate Location Under Complex Background And High Definition

Being a part of LPR, license plate location has a direct influence on the result of character segmentation and recognition. Surveillance of vehicles on the high way is an important scenario for applying LPR. The image under this scenario tends to have complex background and high definition. This thesis studies on license plate location under images which are captured from high way camera.Due to some drawbacks of classic methods on images with high resolution and complex background, this thesis develops an exclusive method to tackle this problem, which satisfies the demond of accuracy as well as efficiency.The developed method contains three aspects of study:Fisrt, the thesis studies a new character which describes the area and Support Vector Machine. An improved color-texture histogram is developed based on HSV color space, which is used as the input of SVM. As the experiment shows, the new character can achieve a better result for image matching.Then the thesis studies the rough detection part of online detection. Due to the drawbacks of Canny and Sobel edge detection algorithm, a new combined method is developed. With the help of filter and projection, the potential plate areas will be extracted. Rough detection based on edge detection can find the region of interest quickly and filter most of the backgournd areas, thus increasing the efficiency of detection using machine learning.Finally, the thesis covers the study of precise detection. A zooming detector is used to scan the candidate band regions. The character in the detector is calculated, and compared with classifiers trained by SVM to tell if the region in the detector is a plate region. Character expension, area ranking, incline correction are introduced as post processing method to make the detection more precise. Through comparing with two classic plate location methods, the experiment shows a promising result of the proposed method.