Automatic Deep Learning-based Temporal Video Segmentation Framework
Tudor Barbu · 2021
A novel automatic deep learning-based temporal video segmentation technique is proposed in this research paper. The considered approach uses machine and deep learning solutions to detect the video shot transitions in a movie sequence. First, a high-level image feature extraction is performed on all the video frames, by applying a pre-trained convolutional neural network. The values of the distances between any two feature vectors corresponding to successive frames are then computed and clustered automatically, in order to determine the visual content discontinuities. Video segmentation experiments and method comparisons are also described here.