Video Shot Boundary Detection using Machine Learning and Confirmation Stage

Shot Boundary Detection or Temporal Video Segmentation is the building block for various video processing application like Video Summarizing, Video Classification, Content-based Video Indexing and Retrieval and many more. So, it becomes crucial to construct a robust Shot Boundary Detection procedure for efficient implementation of video processing systems. In this paper, an efficient dual stage approach to a shot boundary detection procedure has been proposed, which eventually outperforms various contemporary Shot Boundary Detection approaches. firstly, possible transition frames are detected by training a FeedForward Neural Network, whose weights and biases are stochastically optimized by a hybridized Grey-WolfOptimization and Cuckoo-Search algorithm. This process mainly works with the fusion of supervised learning (labelled data) and Reinforcement Learning (Exploration and Exploitation). This results, possible abrupt transition frames generated by this methodology. These frames are further pruned with other features including illumination features to increase the accuracy of the Shot Boundary Detection in terms of precision and F1 Score.