Attention-based deep supervised hashing for near duplicate video retrieval

With the explosive growth of video data on the Internet, near duplicate video retrieval (NDVR) has become an important and challenging issue in the field of information retrieval. Hashing is typically employed to tackle this issue owing to its low memory and fast retrieval speed. Most of the existing video hashing methods directly adopt image hashing methods or perform the frame-pooling strategy, failing to fully explore the spatio-temporal information of videos. In this paper, we propose an attention-based deep supervised video hashing (ADVH) network for NDVR. To capture richer perceptions and acquire more comprehensive video representations, we use a residual network as the backbone and incorporate an attention module to extract spatio-temporal features of videos and motion information between adjacent frames. Moreover, we design a novel pairwise constraint utilizing supervised information to learn compact and discriminative video hash codes. The experimental results on three benchmark video datasets demonstrate that our proposed model outperforms other state-of-the-art hashing methods in retrieval precision.