Interframe Television Coding Using Gain and Displacement Compensation

This paper presents algorithms for predicting luminance changes in successive television frames. The changes can result when objects in a TV scene move or when illumination varies. By a gradient search technique, which seeks to minimize a functional of the interframe prediction error, we estimate two parameters associated with these luminance changes — displacement and gain. Using the estimates of these parameters, we also develop, for interframe coding, adaptive predictors and a segmentor to determine which pels need to be transmitted. We describe several coder variations and compare them by computer simulations using three substantially different scene sequences. For these sequences, gain compensation with improved segmentation reduced the bit rate of a conditional replenishment encoder by 50.7, 11.1, and 39.3 percent. Displacement compensation reduced the bit rate by 61.0, 24.8, and 14.5 percent. Combined gain and displacement compensation reduced the bit rate by 63.4, 32.2, and 44.6 percent.