A two-stage deep learning framework for counterfeit luxury handbag detection in logo images

Counterfeit luxury handbags are a worldwide problem that imposes a huge burden on society and the economy. The logo of a luxury handbag is a good identification point because its content is clear and often difficult to counterfeit. Generally, the authenticity of luxury handbags is identified by senior appraisers, but the number of professional appraisers is in short supply and the cost is high. Recently, convolutional neural networks (CNNs) have been widely used for counterfeit detection. However, directly applying CNNs to identify luxury handbags in logo images is still challenging due to the following two issues: (1) some identification points in logo images are very subtle and (2) both local and global information of logo images need to be considered. To address the above issues, this paper proposes a two-stage deep learning framework for counterfeit luxury handbag detection in logo images. In this framework, multiple object detection models are developed in the first stage, aiming at obtaining the location of each local identification point. In the second stage, an authenticity classification model combining local and global information is established to output the final authenticity prediction result. To verify the effectiveness of the proposed framework, we have constructed and annotated a luxury handbag dataset containing 639 logo images. The results on this dataset show that the proposed framework can accurately detect local recognition points and identify the authenticity of luxury handbags with high accuracy, outperforming the method that directly applies CNNs to classify the whole logo image (i.e., the one-stage model) and existing CNN-based counterfeit detection methods.