If you know the exact dimensions on a coordinate frame, then you would not do "segmentation", just object location: once you know the object location, you would just snip out the relevant portion of the array (or label that portion of the array all with the same label.)
How regular is the placement of the braille characters? For example, if you know the coordinates of one character then by knowing the standard dimensions would you then be able to trivially calculate the location of the next character on the row? Are the columns also regularly spaced?
This also raises to me the possibility that even if the grid of braille characters is exactly predictable, that the image might be titled with respect to the grid. Is that an aspect that you need to be concerned about? If so then you would probably need to start with some code to determine the image angle and then rotate the image back in to alignment using imrotate.
