A transformation filter estimates global transformation parameters between a current frame and a previous frame of a motion video image according to corresponding relative positions of pixels within the two frames. Such estimated global transformation parameters estimate global translation, rotation, and scaling resulting from camera pan, rotation, and zoom. The current frame is transformed according to the estimated transformation parameters. As a result, the transformed current frame more closely correlates with the previous frame and motion compensated compression provide better compression and reduced distortion simultaneously. The transformed current frame is encoded using entropy-constrained vector quantization (ECVQ) using quadtrees for variable block sizes within the representation of the current frame. Such quadtree ECVQ is improved significantly when the single frame is first globally transformed in accordance estimated transformation parameters between the single frame and the previous frame. As a result, larger block sizes can be used more efficiently when differences between the current frame and the previous frame are due primarily to camera pan, zoom, and/or rotation. Of course, when significant differences between the frames are due primarily to motion in the subject matter of the motion video image, smaller frame sizes adapted through ECVQ quadtree encoding still reduce the size of the encoded frame without unduly sacrificing motion video image quality.