Video shot detection in transform domain

Abstract

Content structure is an important aspect in the understanding of video. In this paper, we demonstrate that knowledge about the structure can improve the performance of content analysis operations such as feature extraction, shot transition, shot duration and activity. We have proposed two concepts with the aim to improve the performance of existing Video Shot detection methods. First, we have used a number of Transformations to convert the frames in a video sequence from intensity domain to various other domains. Second, we have used simple algorithms like Pixel Difference and Histogram Difference to the input video sequence of each Transform domain and demonstrate the Shot Detection on a database of Sports clips. The process of Domain Transformation is time intensive and requires high computational resources. Hence it is necessary to find memory handling and process distribution techniques to facilitate Video Shot Detection. In order to handle the seamless video sequence efficiently, we have proposed two different ways of handling the memory. Multithreading and Process Concurrency is the underlying principle employed in both these models. Finally the performance of the Video shot detection method with the best Transform Domain and best Memory model suitable for a real life application is determined.