Aiming at the coding of prediction errors based on the Bayer pattern images

an effective and adaptive arithmetic coder is designed.Generally

a probability distribution of the prediction errors forms Laplacian distribution with zero mean.To reduce computing complexity of estimation of the error probability

the error is classified into quantization regions by minimizing mean square error

and a table is designed which lists the error quantization regions and corresponding probability with 3 significance bits.Experiment result shows that the proposed arithmetic coder is more effective than Golomb coder.