Robust License Plate Detection in Complex Scene using MSER-Dominant Vertical Sobel

摘要

This paper presents a robust method to locate a license plate in a complex scene. In contrast to the most existing method which use the non-handcrafted feature to locate the license plate area, our method uses a modified handcrafted feature. We used Maximally Stable Extremal Region (MSER) combining with dominant vertical Sobel to construct essential biner images. Closing morphology operation is implemented to merge the contour extracted by MSER and dominant vertical edge detection. Based on the area and ratio of contour, license plate candidate area is selected. Furthermore, Support Vector Machine (SVM) is introduced to choose a license plate area by analyzing the Histogram of Oriented Gradient (HOG) of each candidate. For performance evaluation, two datasets consisting of complex scene images under different conditions are tested. The main advantage of this approach is that it is faster than non-handcrafted feature-based method while maintaining the high accuracy of plate detection.