License Plate Detection in Complex Scenes based on Improved Gradient and Match Filtering

Locating the license plate in the image is a key initial step in automatic License Plate Recognition via machine vision. This step becomes more complicated when the image scene is arbitrary (complex) and the view-angle of camera and luminance conditions are not calibrated and constant. In this paper, a new License Plate Detection method in complex scenes is proposed. After removing regions that have a low probability of being a part of plate we apply a series of match filters on the image. These filters correspond to various (unknown) possible sizes of license. Novel improved Gradient operator improves the performance of the filtering stage. Finally a decision is made based on the faithfulness of candidate regions. This step extracts the plate template. This method was applied on 80 gray scale images of complex scene. The positive detection rate of this method was 71% and there was 29% false positive detection.