Spatial Subsampling in Motion-Compensated Television Coders

Television signals contain a significant amount of frame-to-frame redundancy. Interframe coders attempt to exploit this redundancy by (i) Segmenting each television frame into two parts, one part that is predictable from the previous data, and one part that is unpredictable. (ii) Transmitting two types of information: (a) addresses specifying the location of the picture elements in the unpredictable area, and (b) 1895 information (usually quantized prediction error) by which the intensities of the unpredictable area can be updated. (iii) Matching the coder bit rate to the channel rate. Since the motion in a real television scene occurs randomly and in bursts, the amount of information about the unpredictable area will change as a function of time. It is transmitted over a constant bit-rate channel by, (a) storing it in a buffer prior to transmission to smooth out the transmitted information rate, and (b) using the buffer fullness to regulate the encoded bit rate by varying the amplitude, spatial, and temporal resolution of the television signal. Intensities of the picture elements (pels) in the unpredictable areas are transmitted by predictive coding. In conditional-replenishment coding,1"4 quantized values of frame difference, element difference, and line difference (or a combination thereof) are transmitted. In motion-compensated coders,5-9 estimates of interframe translation of objects are obtained, and more efficient predictive coding is performed by taking differences of elements from the previous frame that are appropriately translated.