I am working on some pre-processing for lung CT images. I see a nice tutorial in here. Two of them are normalization and zero centering. I wonder what is the difference between these two steps? If one normalizes the image pixel values to range [0, 1], is it have a benefit to also do zero centering? Is there a difference if I do normalization first or zero centering first?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
3
what is the difference between these two steps?
In that specific Notebook that you linked, normalization means: shrink a numerical distribution in the [0,1] interval. It is commonly referred to as Min-Max Scaling. Shrinking the distribution in the [0,1] interval moves its mean somewhere between 0 and 1.
Zero-centering instead means: to "shift" the values of the distribution so that its mean is equal to 0.
--
if one normalize the image pixel values to range [0, 1], is it have a benefit to also do zero centering?
Zero centering variables improves ML models performance. Activation functions tend to be very responsive to weights' changes especially around zero. I found some good explanations here.
--
Is there a difference if I do normalization first or zero centering first?
The order matters, since "normalization" would move the mean away from zero.
-
$\begingroup$ thank you, your answer is helpful but till now I am confused about the difference, I think normalization is do the zero centering also, isn't it? or I am wrong? $\endgroup$– HunarCommented Jun 22, 2019 at 8:33
-
2$\begingroup$ Normalization, as defined in the Notebook you linked (i.e. Min-Max scaling), is not zero centering. Shrinking the distribution in the [0,1] interval moves its mean somewhere between 0 and 1. $\endgroup$– LeevoCommented Jun 22, 2019 at 9:09
-
1$\begingroup$ OK thank you, could you please add this somewhere in the answer to be accepted as true answer. $\endgroup$– HunarCommented Jun 22, 2019 at 10:20