Concurrent evolution of pixel predictor and context modeling for image coding
Seishi Takamura, Atsushi Shimizu · 2016
Lossless image coding process predicts the value of current pixel from previously decoded pixel values. Then the prediction error is classified according to the context model. This classification splits the sources with different distributions and hence reduce the total entropy of the prediction error signals. In the literature, the predictor has been intensively studied. Some evolutionary approaches have been applied to generate a predictor to improve compression performance. However, the context modelling method has not relatively been well studied. We propose and investigate a novel method to automatically obtain evolved pair of pixel predictor and context modeling. Simulation results show 1.32-3.90% bit-rate reduction against the pair of predictor and context modeler of one of the best conventional methods (CALIC). It is also demonstrated that the evolved algorithm's size is more compact than former results. We also found that context modeler is evolved in more complex form than the predictor.