An Internet-television mash-up application is a new concept for surfing the Internet and watching television simultaneously. This paper introduces one such system, which uses image processing techniques for logo detection and character recognition.

The authors propose a home infotainment platform for Internet-television mash-up applications, which comprises three sub-systems. The first sub-system, television channel identity as context, creates a channel logo template, marks pixels of interest in regions of interest, and matches logos. The second sub-system, textual context from static pages in broadcast television, maps a priori regions of interest; pre-processes, binarizes, and segments touching characters; and automatically detects text using an optical character recognition engine. The third sub-system, textual context from text embedded in broadcast video, localizes suspected text regions, confirms text regions, binarizes, recognizes text, and selects keywords.

For their experiment, the authors observed 92 popular Indian television channels and categorized them into seven classes based on their visibility: (1) 28 are static, opaque and rectangular; (2) 19 are static, opaque, and non-rectangular; (3) 38 are static, transparent background, and opaque foreground; (4) two are static, alpha-blended with video; (5) two are non-static, changing colors with time; (6) two are non-static, fixed animated logos; and (7) one is a non-static, randomly animated logo. This system obtained recall and precision rates of 96 percent and 95 percent for 92 channels, with unrecognition rates of 4 percent and 5 percent due to dislocation and false detection of the channel logos respectively.

The authors encourage Internet-television mash-up applications to bring the Internet to the masses in developing countries and suggest improving poor Internet interactivity.