Images to Signals, Signals to Highlights

Abstract

In this paper, we propose a framework to generate cricket highlights from broadcasted cricket matches. Generating cricket highlights is a difficult problem, due to the duration and rules of the game. We formulate the highlight generation problem as a key-event initialization and key-event-closure identification problem. We propose an Inverse Hierarchical Framework, which is generic and capable of automatically generating highlights of a broadcasted cricket match. We introduce a novel context-aware approach for event-initialization and a Structural Similarity Index-based approach for event-closure detection. Despite the quality of highlights being a subjective measure we provide an evaluation of our framework by comparing it with official highlights on various metrics. We also perform a user-survey on the generated highlights. The approval of the users and overlap between the generated highlights and official highlights indicate the robustness of our framework.