For a real-valued kernel, we would need to re-write the code to generate a real-valued output matrix. However, for proper displaying, the output matrix values would need to be converted to 8-bit unsigned integers via rounding and saturation.

Output Matrix values (generated by a .c code): Output matrix values converted to 8-bit unsigned integers:

278.35
-256.78
-128.59
78.25


I'm not asking for a direct answers, but I'm asking for a clearer interpretation of the sentence "values would need to be converted to 8-bit unsigned integers via rounding and saturation". In plain English this, to me, would be as simple as
278.35 --> 255
-256.78 --> 0
-128.59 --> 0
78.25 --> 78

But I don't think it's that simple. If this value was stored in a double or any other data that is capable of representing this those value, would I just simply snip the MSB and use the 8 LSB? Except this wouldn't really work with floating points...Much confusion.


Enjoy this high res wallpaper. :)