A Deep Video Hashing Algorithm Based on CNNs

This article proposes a new deep video hashing algorithm to improve video retrieval speed. This method utilizes the powerful feature extraction ability of neural networks (CNNs) to extract representative video and temporal features using the spatiotemporal information after stacking convolutional pooling layers, and then calculates compact binary codes. In the loss function, a dual objective loss function is used, which aims to maximize the inter class distance of video feature pairs and minimize the intra class distance, as well as minimize the loss between real valued codes and binary codes. This training process uses the Siamese network. Finally, calculations were conducted on databases CCB and JHMDB, and compared with other hash methods. The experimental results showed the effectiveness of the method.