Abstract: Abstract: Vehicle logo recognition plays a more and more important role in intelligent transportation systems and has attracted extensive attention of researchers. Most traditional VLR methods are based on hand-crafted descriptors for which much heuristic knowledge is required, and thus are hard to adapt to complex and changeable realistic scenarios. Compared with hand-crafted descriptors, the feature learned methods perform betterin solving computer vision problems in complex environments. In the present study, a logo recognition method based on objective optimization learning is proposed to solve the VLR problem in this paper. First, feature parameters are automatically learned from pixel different matrix (PDM) extracted from raw images. Then, the PDMs are mapped into compact binary matrices with the learned feature parameters, and then the codebooks are learned from binary matrices with supervised learning. Finally, the feature vectors are extracted from test images with the learned feature parameters and codebooks. Experiments are carried out on open datasets HFUT-VL and XMU, and the results are analyzed and compared with other state-of-the-art methods. Experimental results show that our method can obtain higher recognition accuracy than hand-crafted descriptor based methods, and less training and testing time is required than deep learning based methods.

Key words: Keywords: vehicle logo recognition, objective optimization, feature learning, codebook, pixel difference matrix