Segmented Optimal Linear Prediction applied to Lossless Image Coding

Most lossless coding techniques use prediction to reduce first order entropy of pixels followed by a data compression scheme like Huffman Coding, Lempel Ziv algorithm or Arithmetic Coding. In this work it is presented a prediction scheme that uses linear coefficients optimized by regions of the set where prediction base pixels lie. This technique was applied to several test images. First order entropies and conditional first order entropies were measured and compared with figures obtained using the well known Median Adaptive Predictor (M.A.P.). It is observed that M.A.P. is a particular case of our scheme with non-optimized coefficients.