At LatinDisplay / IDRC 2012, which was held from November 26 to November 30 at the Universidade Presbiteriana Mackenzie in Sao Paulo, Brazil, coordinated papers from visual scientists on two continents, presented ideas that have interesting ramifications for video-signal compression strategy.

The papers were “The Aesthetic Side of Visual Quality (Judith Redi, Delft University) and “Incorporating Visual Attention Models into Visual Quality Metrics ( Mylene Farias and Wellington Akamine, University of Brasilia). One of the concepts discussed in both papers was “saliency maps. The maps, based on eye-tracking measurements of subjects, show where in an image subjects direct their attention. One example used in the papers (which had been coordinated with each other) was an image of a group of boatmen in a long canoe navigating the rapids in a river. In this complicated image, viewers consistently focused on the faces of the boatmen, and virtually ignored the non-central areas of the image that portrayed the forested riverbank.

Clearly, providers of video content could compress these low-saliency areas with impunity since viewers pay almost no attention to them. The problem is that the most accurate saliency maps are currently created with painstaking research on human subjects. There are computational models that generate approximations of the experimentally derived saliency maps, but the different models produce maps that differ significantly from each other. Although they are different, could some version of these models be accurate enough to compress video data streams more efficiently and (we hope) less objectionably? We would need to predict the saliency of image areas in real time and use that information compress the pixel map selectively. That's a roadmap that could begin driving development strategies in the foreseeable future.

Ken Werner is Principal of Nutmeg Consultants, specializing in the display industry, display manufacturing, display technology, and display applications. You can reach him at [email protected].