Real-time license plate localization based on a new scale and rotation invariant texture descriptor

In this paper, we present a real-time and robust license plate localization method for traffic control applications. According to our approach, edge content of gray-scale image is approximated using line segments features by means of a local connective Hough transform. Then a new, scale and rotation invariant, texture descriptor which describes the regularity, similarity, directionality and alignment is proposed for grouping lines segments into potential license plates. After a line-based slope estimation and correction, false candidates are eliminated by using geometrical and statistical constraints. Proposed method has been integrated in a optimal license plate localization system. Evaluation is conducted on two image databases which were taken from real scene under various configurations and variability. The result shows that our method is real-time, robust to illumination condition and viewpoint changes.