bias setting in generalconvolution

Wilbert · 1st December 2003, 22:57

Does someone know what the bias setting is generalconvolution is doing? Is it optional?

Richard Berg · 2nd December 2003, 03:57

If I recall, the general technique is to perform the convolution, divide by the divisor, and then add the bias.

Looking at the code, apparently a divisor isn't implemented. That's a shame considering the matrix elements have to be integers. I can add that to the to-do list.

Wilbert · 3rd December 2003, 21:33

Quote:

If I recall, the general technique is to perform the convolution, divide by the divisor, and then add the bias.

Could you say a little bit more, because it still sounds vague? Why would you use a different bias as zero?

Richard Berg · 4th December 2003, 05:55

It's nothing more than "turning up the volume." This is often used when a convolution will result in both negative and positive values, and you want to see the full range before it's clamped at 0. Emboss filters work this way.

1st December 2003, 22:57	#1 \| Link
Wilbert Super Moderator Join Date: Nov 2001 Location: Netherlands Posts: 6,392	bias setting in generalconvolution Does someone know what the bias setting is generalconvolution is doing? Is it optional?

2nd December 2003, 03:57	#2 \| Link
Richard Berg developer wannabe Join Date: Nov 2001 Location: Brooklyn, NY Posts: 1,211	If I recall, the general technique is to perform the convolution, divide by the divisor, and then add the bias. Looking at the code, apparently a divisor isn't implemented. That's a shame considering the matrix elements have to be integers. I can add that to the to-do list.

4th December 2003, 05:55	#4 \| Link
Richard Berg developer wannabe Join Date: Nov 2001 Location: Brooklyn, NY Posts: 1,211	It's nothing more than "turning up the volume." This is often used when a convolution will result in both negative and positive values, and you want to see the full range before it's clamped at 0. Emboss filters work this way.