Compressed bitrate analysis project.

Data-driven modeling of video compression

Mentor: Roger Donaldson, Midvale Applied Mathematics


Despite our hand in their design, and hence their seemingly deterministic operation, electronic systems grow ever more complex and are often most simply characterized through measurement-driven modeling. This workshop presents measurements of the rate at which bits are produced as output of a compression algorithm for a digital video camera. The data are produced by presenting the camera with a variety of test scenes, and operating it over a variety of settings. The data presented here can both simulate the camera’s bit rate as a function of operating conditions for design purposes, as well as provide some insight into how the camera’s compression algorithm operates.

Several other goals will be presented, and participants will be free to investigate a variety of possible directions with the data. Perhaps most importantly, complete details of how the data are generated — namely the internal workings of the camera’s algorithm — are not available. Despite this fact, we believe that systems as elaborate as those that can control the camera’s bit rate under changing conditions can be achieved with good modeling practice on these available data.