Spatial Feature Collaborative Network for Trademark Image Retrieval

Trademark is the mark of the brand and widely used in various types of advertising media and e-commerce sales. As the number of trademarks grows at high speed, efficient trademark image retrieval and management become more and more critical. Traditional trademark retrieval models introduce low-level visual, textual, and labels information for retrieval, but are of relatively poor performance. Recently, benefitting from the Deep Learning model, content-based image retrieval makes excellent progress. Typically, the trademark image contains sufficient semantic visual information. In this paper, we proposed a model named Spatial Feature Collaborative Network (SFCN) for trademarks image retrieval. SFCN fuses AlexNet and Faster R-CNN to extract global and region features, which are used for global and local collaborative retrieval. We conducted extensive experiments on a specific trademark dataset, and the experimental results show that our proposed method achieves better performance than other baseline models. Moreover, our proposed framework has been running in a real-world environment.