Transformed Deep Spatio Temporal-Features with Fused Distance for Efficient Video Retrieval

For the goal of video retrieval, this research proposes wavelet transformations on deep spatiotemporal characteristics. The component-wise similarities between the query video feature and prototype video feature are calculated because level 1 wavelets extract two components from any signal or feature vector. The ultimate dissimilarity for determining the top 1 and top 5 accuracy is created by fusing these differences. The outcomes demonstrate that the suggested technique performs better than a baseline strategy. The following strategy for improvement can be investigated further by employing fast learning networks that are trained on the training sets of both data sets to provide better classification of the query as well as the prototype feature vectors, which would enhance the retrieval accuracy.