Adding Z-Normalization before or after RandomNoise #843
-
|
Hey, I have a question regarding the Z-Normalization and its position in the pipeline w.r.t RandomNoise. According to the online Colab, the Z-Norm should be applied before adding RandomNoise. However, if we normalize the image to zero mean and unit variance and add noise with varying mean/variance, wouldn't that alter our mean and variance of the image again, making the Z-Norm "obsolete"? Thank you for your help :) |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
|
Hello |
Beta Was this translation helpful? Give feedback.
-
|
Hey, It is in the Data and Augmentation Tutorial with a link to the Random Noise cell. If the above link to the specific cell is not working for you, use the link to the regular tutorial , navigate to Intensity Transforms > Augmentation > Random Noise.
I am just wondering about the sequence of operations as RandomNoise before Z-Normalization would make more sense to me - unless one adds noise with zero mean and unit variance if used after the normalization (which however would make the parameters obsolete as only one configuration can be used) Thank you for your help - maybe I am confusing some things here :) |
Beta Was this translation helpful? Give feedback.
-
|
As explain the point here is that adding noise to an image it is difficult to choose a "relevant" std for the noise, that is why "normalising the intesity" first is important. to be able to keep a given std (for RandomNoise). Note that you could also do it with the if Z-Normalization is important, you can just add it a second time after adding random noise |
Beta Was this translation helpful? Give feedback.
As explain the point here is that adding noise to an image it is difficult to choose a "relevant" std for the noise,
(imagine you have intensity mean around 200 in one subject and 2000 in an other (that can append for MRI because it is always define with an arbitrary scale factor)
so adding a noise with a std of 10 will have very different effect of those 2 subjects
that is why "normalising the intesity" first is important. to be able to keep a given std (for RandomNoise). Note that you could also do it with the
RescaleIntensitytransformif Z-Normalization is important, you can just add it a second time after adding random noise