Static Object-of-Interest Recognition from Video Streams for Effective Visualization in Small Hand-Held Displays

摘要

Multimedia devices are becoming smaller to enhance mobility, while users' demands for better Quality of Experience (QoE) are also increasing. In this paper, we present a method to increase QoE in small hand-held devices during the enjoyment of sports videos. Static objects are superimposed graphics such as score boxes, expository comments, or logos that are attached during transmission by broadcasters. Among them, viewers are interested in the graphics that holds important information, called a static object-of-interest (OOI). Due to the resolution and size limitation, such information may not be clearly understood in small display devices. For example, the static OOIs like tiny score-boxes can hardly deliver the details of games to viewers. To cope with this problem, we propose a real-time recognition method for static OOIs in sports videos to visualize the OOIs effectively in such small displays. First, various static objects are detected automatically using the edge map analysis, and then, those objects are classified through supervised learning to determine static OOIs. Finally, the static OOIs, after its size and location are adjusted to fit a small-sized display, are attached to the original video to provide mobile users with a better viewing experience. Experimental results show that the proposed method works effectively (97% accuracy on average in recognizing OOIs) and efficiently (0.25 seconds on average from detecting static objects to browsing), which is sufficient for real-time viewing. The application context addressed in this study is sports videos but other application areas also can benefit.