DICOM image compression using a hierarchy of predictors

We present a new method for lossless image compression in the domain of medical imaging. The core of this method is the use of a hierarchy of predictors. In particular, for each examined pixel, a context based on appropriate quantization of neighboring pixel deltas is determined. A different predictor is used for each context allowing for a far more accurate prediction. The contexts form a hierarchy based on increasingly refined quantization of deltas. A number of variations on this basic approach are evaluated and compared with JPEG-LS.