Spatial domain decomposition for coding image sequences

Summary form only given. A spatial technique for the efficient adaptive coding of image sequences at low data rates is reported. The technique has the advantage of simple implementation for real-time operation (avoiding the use of floating-point arithmetic) and allows the matching of coding resolution to variations in image detail. The basis of the technique is a recursive quadtree division of image frames into successively smaller and smaller blocks until the original picture detail can be adequately represented by a bilinear interpolation from the four block corner points. At this point subdivision ceases and information is transmitted about the structure of the quadtree division and any new corner points needed for interpolation (for most blocks, some corner points will have already been coded to reconstruct previous blocks). The coding algorithm can process sequences on an adaptive intra- or interframe basis, the latter employing conventional motion compensation techniques; the inclusion of the former capability reduces the 'dirty window' effect that usually accompanies conditional replenishment type coding of the interframe difference signal alone. Initial problems with the application of the algorithm to an image sequence derived from the effects of noise causing random failure of the threshold test and errors in corner point values, which are then spread over the whole block by the reconstruction interpolation. Methods of counteracting these effects as well as techniques for optimizing the quantization of sample values, particularly in the case of the processing of small blocks, have been developed.