Accurate multi-scale license plate localization via image saliency

In modern society, the automatic vehicle license plate (VLP) detection has been widely used in the fields of intelligent transportation system (ITS) in order to build an effective automatic traffic management system. As VLPs are salient to human visual system, we propose a novel method for VLPs detection based on saliency mechanism. The proposed method mainly contains three parts. The first part is candidate detection combining both colour and gradient information. The second part is affine rectification by conducting rotation and shear transformation to rectify tilt plates. The third part is accurate location of plates using saliency of the proposed candidates in the second part. Experimental results on a large set of vehicle images and street-view panoramas demonstrate that our proposed method can not only accurately locate the license plates with different background/ foreground colours at different inclination angles and different scales but also achieve a higher true positive rate and a lower false positive rate with a low computational cost.