# Stacking Z dimension images using SimpleITK

**URL:** <https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623>\
**Category:** Beginner Questions\
**Tags:** simpleitk, python\
**Created:** [May 9, 2024, 12:12pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623 "2024-05-09T12:12:00Z")\
**Posts on this page:** 7\
**Page:** 2

<div class="post-metadata">

**Author:** ![dzenanz](https://discourse.itk.org/user_avatar/discourse.itk.org/dzenanz/32/1093_2.png) [@dzenanz](https://discourse.itk.org/u/dzenanz)\
**Post date:** [May 10, 2024, 6:19pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/21 "2024-05-10T18:19:40Z")

</div>

You probably need to apply flat-fielding. C++ example is [here](https://github.com/InsightSoftwareConsortium/ITKMontage/blob/79c81953d87f891f9a2c96457480bdc9f9110f24/examples/CompleteMontage.cxx#L149-L317). The key class for that, `N4BiasFieldCorrectionImageFilter`, is wrapped in Python. However, that is best done via hardware-specific way, e.g. acquire “dark” and “white” images and use them for flat-fielding.

---

<div class="post-metadata">

**Author:** ![sleepingcat4](https://discourse.itk.org/user_avatar/discourse.itk.org/sleepingcat4/32/4032_2.png) [@sleepingcat4](https://discourse.itk.org/u/sleepingcat4)\
**Post date:** [May 10, 2024, 8:28pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/22 "2024-05-10T20:28:48Z")

</div>

I used a similar method like you suggested and it resulted in no improvement.

Code:

```auto
import SimpleITK as sitk #1

raw_img = "resampled_image.tif"

raw_img_sitk = sitk.ReadImage(raw_img, sitk.sitkFloat32) #2
raw_img_sitk_arr = sitk.GetArrayFromImage(raw_img_sitk) #3

transformed = sitk.RescaleIntensity(raw_img_sitk, 0, 255) #1
transformed = sitk.LiThreshold(transformed,0,1) #2
head_mask = transformed

shrinkFactor = 4 #1
inputImage = raw_img_sitk

inputImage = sitk.Shrink( raw_img_sitk, [shrinkFactor] * inputImage.GetDimension() ) #2
maskImage = sitk.Shrink( head_mask, [shrinkFactor] * inputImage.GetDimension() ) #3

bias_corrector = sitk.N4BiasFieldCorrectionImageFilter() #4

corrected = bias_corrector.Execute(inputImage, maskImage) #5

log_bias_field = bias_corrector.GetLogBiasFieldAsImage(raw_img_sitk) #1
corrected_image_full_resolution = raw_img_sitk / sitk.Exp( log_bias_field ) #2

out_path = "field_corrected_image.tif"

# Cast the pixel type to float
corrected_float = sitk.Cast(corrected_image_full_resolution, sitk.sitkFloat32)

out_path = "corrected_resampled_image.tif"
sitk.WriteImage(corrected_float, out_path)

```

![image](https://discourse.itk.org/uploads/default/original/2X/9/93e33e1c6056789590933b2a31cca9655a5823a2.jpeg)

I was reading through Optical flow to fix “artifacts” that comes-up after tile merging and it looked nice. You have any thoughts or other algorithms to point to? [Registration using optical flow — skimage 0.23.2 documentation](https://scikit-image.org/docs/stable/auto_examples/registration/plot_opticalflow.html) @dzenanz

---

<div class="post-metadata">

**Author:** ![dzenanz](https://discourse.itk.org/user_avatar/discourse.itk.org/dzenanz/32/1093_2.png) [@dzenanz](https://discourse.itk.org/u/dzenanz)\
**Post date:** [May 11, 2024, 1:21am UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/23 "2024-05-11T01:21:11Z")

</div>

In my experience, `N4BiasFieldCorrectionImageFilter` needs a decent estimate of what is tissue and what is background in order to work well.

---

<div class="post-metadata">

**Author:** ![sleepingcat4](https://discourse.itk.org/user_avatar/discourse.itk.org/sleepingcat4/32/4032_2.png) [@sleepingcat4](https://discourse.itk.org/u/sleepingcat4)\
**Post date:** [May 11, 2024, 4:06pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/24 "2024-05-11T16:06:30Z")

</div>

do you have any advise how to provide that decent estimate? @dzenanz

---

<div class="post-metadata">

**Author:** ![dzenanz](https://discourse.itk.org/user_avatar/discourse.itk.org/dzenanz/32/1093_2.png) [@dzenanz](https://discourse.itk.org/u/dzenanz)\
**Post date:** [May 13, 2024, 12:21pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/25 "2024-05-13T12:21:33Z")

</div>

I ran into that with MRIs of the head. I had an iterative approach: threshold to get some segmentation of the head, apply bias field correction using that segmentation, then threshold the corrected image and iterate that a few times.

Your case looks harder, I don’t think this approach would work for you.

---

<div class="post-metadata">

**Author:** ![zivy](https://discourse.itk.org/user_avatar/discourse.itk.org/zivy/32/1726_2.png) [@zivy](https://discourse.itk.org/u/zivy)\
**Post date:** [May 13, 2024, 12:43pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/26 "2024-05-13T12:43:42Z")

</div>

Hello @sleepingcat4,

This looks like [vignetting](https://en.wikipedia.org/wiki/Vignetting). Possibly look at the [PyBaSiC](https://github.com/linum-uqam/PyBaSiC) tool to deal with these artifacts in Python, also available as [a Fiji plugin](https://sites.imagej.net/BaSiC/).

---

<div class="post-metadata">

**Author:** ![blowekamp](https://discourse.itk.org/user_avatar/discourse.itk.org/blowekamp/32/79_2.png) [@blowekamp](https://discourse.itk.org/u/blowekamp)\
**Post date:** [May 13, 2024, 1:04pm UTC](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623/27 "2024-05-13T13:04:59Z")

</div>

I agree it looks like vignetting. This should be corrected on a per tile basis before composing the mosaic.

The contribution of vignetting to the image can be estimated from images of a constant or empty field. It may even be constant between images.

[Previous page](https://discourse.itk.org/t/stacking-z-dimension-images-using-simpleitk/6623.md?page=1)
