To train a backprop neural net to recognize a playing card viewed from an iPad, I want to get the pixel counts of the training set card image and the actual card image to correspond as much as practical by controlling how far the real playing cards are from the iPad when real recognition is attempted. I am attempting to train the network on artificially generated images because such generation is much easier, I think.
I don't quite understand how to translate the camera lens' quality to pixel count of the training set. My iPad produces .jpg pictures of "Dimension"720 x 960 (which for the example photo I used for this messages is 227 KB and is not a playing card). When I open that picture in Photoshop Elements it says that at 100% (RGB/8) the image is of size 10x13.333 inches and 72 ppi. A playing card is 2.5x3.5 inches. So can I get some guidance on whether that is enough information and if so, how to use the information? Thanks, -- (B=) ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
