Lossless compression for medical imaging systems using linear/nonlinear prediction and arithmetic coding

A differential pulse code modulation (DPCM)-based system is proposed for lossless compression of medical images. Three techniques are used in the system. The intensity of the current pixel is predicted optimally by linear combination of the intensities of N pixels in a neighborhood of the causal region. The multilayer perceptron, a feedforward neural network, is used to model the nonlinearity in images. A hierarchical prediction scheme is adapted for progressive transmission. Coefficients for each hierarchical layer are chosen optimally to achieve minimum entropy. Prediction errors, assumed to be Laplacian distributed, are encoded by the Q-coder, an adaptive arithmetic coder