140 - A Novel Approach for Fast and Robust Multiple License Plate Detection
License Plate Detection (LPD) is the most difficult,critical and time consuming task in license plate recognition (LPR) systems. In this paper, a novel texture-based method is proposed for fast and robust LPD. First, a new filter called Peak-Valley filter is applied on the lines of the image. This filter not only extracts the remarkable gray level changes as consecutive peaks and valleys, but also simultaneously removes the undesirable small variations. Secondly, a sequential Peak-Valley partitioning is utilized to segment the transitions into some groups. Afterward, a neural network is employed to find true candidate lines and finally the candidate lines are aggregated to form the plates regions. According to our experiments, the proposed method correctly detects all plates presented in the image regardless of their styles and without considering the whole image. Experimental results showed that this approach can apply on real-time application for outdoor complex scenes.
|