Maximum entropy model-based baseball highlight detection and classification

In this paper, we propose a novel system that is able to automatically detect and classify baseball highlights by seamlessly integrating image, audio, and speech clues using a unique framework based on maximum entropy model (MEM). What distinguishes our system is that we emphasize on the integration of multimedia features and the acquisition of domain knowledge through automatic machine learning processes. Our MEM-based framework provides a simple platform capable of integrating multimedia features as well as their contextual information in a uniform fashion. Unlike the Hidden Markov Model and the Bayes Network-based approaches, this framework does not need to explicitly segment and classify the input video into states during its data modeling process, and hence remarkably simplifies the training data creation and the highlight detection/classification tasks. Experimental evaluations demonstrate the superiority of our proposed baseball highlight detection system in its capability of detecting more major baseball highlights, and in its overall accuracy of detecting these major highlights.