An EMD based approach for Saliency Detection in Multimedia Data

The advancement of multimedia technology has increased the problem of its storage, transmission and display etc. As this multimedia data is mostly targeted to humans, multimedia researchers are working in field of incorporation of human's perception, action related abilities. Detection of interesting content in images or videos based on human's attention-based perspective is termed as Saliency. Various existing popular methods which have been used for saliency estimation were based on techniques like Difference of Gradients (DoG), K1 Divergence. These methods are time consuming and are applicable on selected parts of input only. This causes need to find a better approach which should be time-efficient and provides better results. In this paper Earth Mover's Distance (EMD) based approach is adopted with bottom-up features-based saliency estimation for image and video data. Result shows that incorporation of EMD which is based on calculating histogram differences to find minimal cost incurred in transforming input data as close as output data, provides better results in terms of time consumption.