Deep triplet hashing network for case-based medical image retrieval

Deep hashing methods have been shown to be the most efficient approximate nearest neighbor search techniques for large-scale image retrieval. However, existing deep a poor small-sample ranking performance case-based medical The top-ranked images in returned query results may as different class than image. This problem is caused by classification, regions of interest (ROI), and information loss space. To address problem, we propose an end-to-end framework, called Attention-based Triplet Hashing (ATH) network, learn low-dimensional hash codes that preserve ROI, information. We embed spatial-attention module into network structure our ATH focus on ROI aggregates spatial feature maps utilizing max-pooling, element-wise maximum, mean operations jointly along channel axis. triplet cross-entropy can help map classification similarity between codes. Extensive experiments two datasets demonstrate proposed further improve retrieval compared state-of-the-art boost small samples. Compared other methods, enhance code-discriminability