Hi, I want to use a neural network to determine which character is shown in a bitmap. I take these bitmaps from digital screenshots and I got them in b/w.
I wonder, what dimensions these character bitmaps should have. Right now I'm using the top and bottom of the line, and the left+1 and right+1 bound of each charakter. I use these bounds, as after I took the exact charakter bounds (instead of top and bottom of the row), for example the "," sign was almost just a black bitmap, the same as the "-","l" or "I".
Now which bounds do you think are good for the use of neural networks?
And I got another question, regarding the normalization of a bitmap. I got a widely range of small bitmaps, something like 3x11 to 9x5.. After I set them for example to 20x20, the original look of these bitmap is heavily changed. Can a neural network work with these transformatted images, or should I try to preserve the original shape of a charakter?