Abstract
In predictive coding, prediction methods that aim to more efficient prediction by switching some predictors based on local index have been proposed in the past. Difference between adjacent pixels, i.e. gradients, is popularly used in such methods. However, relationship between the gradients and predictors is discussed only qualitatively. In this paper, we propose a prediction method using difference to average prediction value as local index. In the experiments, the proposed prediction scheme shows its superiority to LOCO-I and CALIC in terms entropy of prediction error and code length.
