IMAGE AND VIDEO CODING

We are one of the few groups worldwide researching "pure" predictive coding of images and video (that is, where predictions are made pixel-by-pixel without an initial block or subband transform). Predictive coding adapts well to image type and compression requirements, which allowed us to develop first Binary Tree Predictive Coding (BTPC) then Adaptive Prediction Trees (APT) [*] as high-performance general-purpose still-image compression schemes. We developed video versions of these, including a method designed for coding of arbitrary computer-screen video [*]. We have also applied APT to 3D-image coding [*].

Click here for the APT homepage.

In object-based coding our most recent work [*] [*] was in motion segmentation and region description. The picture below shows the segmentation derived by global-to-local motion estimation with the coding-by-asymmetric-subdivision structure overlaid.