Abstract: A discrete transform image data compression system in which frequency transform coefficients are modified in accordance with a matrix of quantizer values employs a predefined plurality of quantization matrices to adaptively select, on a document-by-document basis, an approximate memory packet size for each document's compressed image data storage by selecting one of the plurality of quantization matrices in accordance with the packet size estimate obtained for each document image. Additionally, the system employs generation of contrast reduction and gray level stretch remapping curves as a function of global image data characteristics, such as a gray level histogram of the document image data. The remapping curves are utilized to preprocess the image data for more effective data compression.