Unsupervised Approach for Retrieving Shots From Video
Acquiring the video information based on user requirement is an important research that attracts the attention of most of the researchers today. This paper proposes an unsupervised shot transition detection algorithm using Auto Associative Neural Network (AANN) for retrieving video shots. The work further identifies the type of shot transition, whether abrupt or gradual. Key frames are extracted from the detected shots and an index is created using k-means clustering algorithm for effective retrieval of required shots based on user query. The approach shows good performance in retrieving the shots, tested on five popular genres.