Input Encoding

• How encode the image?
• Extract local features: edges, regions of uniform intensity - how handle a varying number of features per image?
• encode image in 30x32 pixel intensity values, with one network input per pixel- pixel intensities were linearly scaled from 0 to 255 down to 0 to 1, why??
• the 30x32 grid is a coarse resolution summary. Coarse pixel intensity is a mean of the corresponding high pixel intensities - reduces computational demands while maintaining sufficient resolution to correctly classify images - same as ALVINN except there a represented pixel chosen at random for efficiency