Abstract: An interband prediction method for image data decomposition and reconstruction in which a coder decomposes image data using pyramid subband image data decomposition. The resulting DC subband is further decomposed and is used to generate first predictors for various allowed geometries. The method then generates a predictor set for each pixel in the decomposed image in which the predictors for each pixel are ranked. The method further generates a message identifying the predictor in each predictor set to be used in the reconstruction of the data. In the reconstruction of the image data, the DC subband image data is used to generate a second predictor set identical to the first predictor list. The desired predictor for each pixel identified by the messages is used in conjunction with the DC subband image data to generate estimated next-level DC subband image data. This process is repeated until the estimated next-level DC subband image data is a predicted estimate of the original image data.