Enhancement algorithm for nonlinear context-based predictors

The authors propose a bicandidate algorithm (BCA) to enhance the prediction accuracy of nonlinear context-based predictors, such as the MED predictor (used by LOCO-I/JPEG-LS) and the GAP predictor (used by CALIC). The BCA provides two predictive values to be selected (i.e. there are two candidates), and only part of the selection should be indexed. To test the performance of the BCA, it is applied to the enhancements of the MED predictor and the GAP predictor. According to experimental results, both the enhanced predictors perform better in prediction accuracy (evaluated by the first-order entropy) at an average improvement rate of 2.8%, and the enhanced MED predictor outperforms the modified MED predictor proposed by Jiang et al. (2000) and the ´soft´ predictors proposed by Estrakh et al. (2001). The predictors enhanced by BCA remain at similar complexity levels and the application of BCA is relatively simple.