Robust license plate detection using profile-based filter

The complex illumination such as extreme dark or light environment has always been a problem for license plate detection algorithm. For the complex illumination problems of most existing algorithms when processing a variety of vehicle images captured from extreme dark to light scenes, we proposed a profile-based filter (PBF) to precisely localize the license plates under different conditions. Meanwhile, we collected a dataset named PKU extreme dataset including 171 extreme dark and light vehicle images for performance evaluation, together with public PKU vehicle dataset including 3977 vehicle images. The algorithm can not only overcome the missing detection caused by strong or dark light but also can keep the parameters independent from the illumination environment. Experiments demonstrate that the filter performed well on both datasets mentioned above with a state-of-the-art average detection rate.