Image processing requires large amounts of memory (main memory and mass storage) and important processing power. Digital signal processors (DSP) can increase processing speeds, but they don't provide any facilities for multiprocessing and for the management of mass storage devices. The present project requires the processing of large aerial photographs (20000×20000 pixels) for the automatic computation of terrain elevation. Currently, a CCD camera scans image parts of 512 by 480 pixels and transmits them to the image processing array. A processing array based on transputers was chosen for the parallelization of image processing operations. Such an architecture can easily be expanded if more processing power and storage capacity are required. The authors describe the message passing system ensuring communications between any processors of the network and several image processing algorithms running in a network which currently includes twenty transputers