I am currently extracting letters from an image to write a letter recognizer using a convolutional neural network. The information I get from the image of text is the position of each letter on the image itself. What I am trying to do is determine where a word ends so I can insert a space when I am printing my result. A line break is quite easy since the distance from the current letter to the next letter is negative and that's how I determine a line break. Now to determine if there is a space between 2 letters doesn't seem that easy.
Things I have tried:
Set a threshold (30 pixels)
Calculate a mean of all distances excluding negative ones
Is anyone aware of a standardized way to determine if there is a space between 2 letters on an image of text? (This should be independent of font and such)