Video Fingerprinting via Deep Metric Learning

In order to improve the compactness,an end-to-end video fingerprinting via deep metric learning is proposed while ensuring its robustness and distinctness.The whole framework is composed of weight-sharing triplet networks.The improved 3D residual network is employed to be the main branch,which fuses multi-layer features together and compresses it.This process maps the raw data to compact fingerprints directly.The new designed boundary-constrained triple angle metric loss and classification loss compose the objective function.The new triple loss overcomes deficient expression to feature correlation.The classification loss function remedies the metric loss which is not sensitive to the overall distribution of sample features.A large number of experiments have been carried out on the FCVID set for the proposed algorithms,traditional methods and deep learning methods.The results show that the algorithm enhances compactness significantly while improving the robustness and distinctness simultaneously.