Using genereated NN to analyse an image

Sep 22, 2010 at 10:40 AM

Hi everyone,

how do you actually parse the image, which is supposed to be analyzed, as a double[]?

Everything else worked out fine, including the learning etc.

I would be grateful for any help on solving this problem...

Thanks in advance, nRg2410

Jan 20, 2011 at 8:09 AM

If you want to analyze an image with more characters (a page), you must first extract the individual characters (look for document layout analysis and segmentation in OCR). A single character image must be resized to fit into the network (29*29) and converted to double[] (if you use the CNN class). Pixel values should be the same as in your training set, in the sample code they range from -1 for black to 1 for white.

Hope this helps somebody.