The coding of pictorial data

We are concerned with the problem of designing an efficient general method of coding two-level pictorial data. Both exact and approximate coding techniques are illustrated. A pilot experiment is presented, in which a digital computer was used to realize two-dimensional predictive coding. Although the resulting compression was not great, there are reasons for believing that this procedure would be more successful with realistic pictorial data. Further experiments, which made use of approximation methods are described. These methods arose from the application of pattern recognition theory to the present problem. Their use, either independently or prior to predictive coding, yielded compression significantly greater than that attained by predictive coding alone.