Component-Based Attention for Large-Scale Trademark Retrieval

The need for large-scale trademark retrieval (TR) systems has significantly increased to combat the rise in international infringement. Unfortunately, ranking accuracy of current approaches using either hand-crafted or pre-trained deep convolution neural network (DCNN) features is inadequate deployments. We show this paper that TR can be improved by incorporating hard and soft attention mechanisms, which direct critical information such as figurative elements reduce given distracting uninformative text background. Our proposed approach achieves state-of-the-art results on a challenging dataset.