Bayesian fuzzy clustering and deep CNN-based automatic video summarization

The expansion of growth in the generation of video data in various organizations causes an urgent requirement for effectual video summarization methods. This paper devises a novel optimization-driven deep learning technique for video summarization. The aim is to give an automated video summarization. Initially, the video data is extracted from the database. Then, the representative frame selection is done using Bayesian fuzzy clustering (BFC). After that, the frames are then temporally segmented, wherein each segment is modelled as a representative frame, which is generated by clustering the temporal segment into clusters. These segments are selected from each cluster closest to the cluster center. The next step is fine refining that is performed using Deep convolution neural network (Deep CNN), which helps to refine the final frame set. The Deep CNN is trained using the proposed Lion deer hunting (LDH) algorithm. The LDH algorithm is the integration of the Deer hunting optimization algorithm (DHOA) and Lion optimization algorithm (LOA). Thus, the final frames obtained by the proposed LDH-based Deep CNN are employed for video summarization. Here, the final frames are adapted to play as a continuous output video. The developed LDH-based Deep CNN offered enhanced performance than other techniques with the highest precision of 0.841, highest recall of 0.810, and highest F1-Score of 0.888.