关键词: 视频指纹, 非局部模块, 3D残差网络, 三元组损失, 量化误差损失

Abstract:

In order to realize fast and accurate retrieval of video copies, this paper proposes a compact video fingerprint based on non-local 3D residual network. Based on the triplet network architecture, the algorithm uses the non-local block 3D residual network to simultaneously capture the global and local spatio-temporal information of the video, and adds the quantization coding layer at the end of the feature extraction part to realize end-to-end mapping of raw video data to discrete fingerprint codes. A large number of experimental results show that compared with the comparison algorithm, the algorithm is outstanding in terms of robustness and uniqueness while maintaining compactness, and the precision and recall rate are significantly improved.

Key words: video fingerprint, non-local block, 3D residual network, triplet loss, quantization error loss