The Research Of Vehicle Logo Recognition Method Based On Visual Features

Vehicle logo recognition(VLR)is a very important part of vehicle Manufacturer and Model Recognition(VMMR),which is widely applied for traffic management,traffic monitoring and traffic crime solving etc.VLR also belongs to image classification area,which has broad application prospects.Currently,existing VLR methods are mostly based on combination of features and classifiers,which don't focus on analyzing the characteristic of vehicle logo images,and then limits the recognition performance.In practical application scenarios,VLR is easily affected by illumination variation and low image resolution.To solve these problems,based on structure information and component information of vehicle logo images,two methods are proposed,vehicle logo recognition based on randomly sampled pixel-pair feature from foreground background skeleton areas and vehicle logo recognition based on OE-POEM feature.Pixel-pair feature can reflect the difference from foreground and background of vehicle logo image,and OE-POEM feature can represent edge information and component information of vehicle logo image,so these two features are suitable for describing vehicle logo.Many experiments have proven these two methods can obtain good recognition performance.The work of this dissertation is mainly listed as follows:(1)Existing vehicle logo recognition methods are concluded and analyzed: existing methods are classified,several representatives of which are introduced.(2)Vehicle logo recognition method based on pixel-pair feature is proposed: combined with the structure information of vehicle logo images,a new pixel-pair feature is proposed,which reflects the difference between foreground and background of vehicle logo image.The experiment results show this method can achieve good recognition performance on low-resolution test samples.(3)Vehicle logo recognition method based on OE-POEM feature is proposed: combined with the edge information and component information of vehicle logo images,a new OE-POEM feature is proposed,which uses gradient information of all orientation.The experiment results show this method obtain high recognition rate on our vehicle logo datasets.