The demo clips

The links marked 'demo' in our MPEG-1 compression resuls point to composite movies that show a combination of input, output, saliency map and internal blurring mask in the arrangement shown above. Although all computations are performed at the original framesize (640x480), the combination movies have been downscaled by a factor two for display purposes.

• Top-left: Original - The original frames.
• Top-right: Multi-Foveated - The foveated results; when discrete foveas are used (as opposed to a continously graded blur), their centers are marked by yellow squares, with the largest square size for the most salient fovea, and smallest square size for the least salient. If a fovea is pointed to a location over 20 times less salient than the most salient location, it will not be used nor displayed.
• Bottom-left: Saliency Map - This is the raw saliency map as computed by our neurobiological model of visual attention. Brighter values indicate higher saliency.
• Bottom-right: Foveation Mask - The foveation mask is a sliding temporal weighted average either of saliency maps (when continuous graded blur is used) or of chamfer distance maps to the objects fixated by the various foveas. Brighter values indicate larger distance and hence more blur; black regions will be preserved without any blur.