There are two aspects of the requirement that can produce great economies of storage (perhaps).
The images are from several different stages of the production process, and it would be well to start with the images from a perfect product and then XOR the corresponding images from the observed processes. If the process is identical then the Result of the XOR will be zero, and anyway it's likely to be more compressible than raw images or successive images. The second aspect of this approach is that a count of the XORed image will indicate if it is far enough from the archetype to be worth saving. Just throw away the similar ones. Be sure to match the image size with what you are looking for. If it's tiny scratches then you need hi res but the eyeball cameras can see a big error almost as well as the more expensive cameras. Yours Sincerely John John Brohan National Instruments LabVIEW expert in Montreal Traders Micro "We connect all sorts of things to computers" 317 Barberry Place DDO Montreal PQ Canada H9G 1V3 Tel (514)995-3749 [EMAIL PROTECTED] http://www.TradersMicro.com/Index.htm