Addition of Image Quality Metrics - Githubissues

TomographicImaging / CIL

A versatile python framework for tomographic imaging

https://tomographicimaging.github.io/CIL/

Apache License 2.0

97 stars 43 forks source link

Addition of Image Quality Metrics #1911

Open lauramurgatroyd opened 2 months ago

lauramurgatroyd commented 2 months ago

Description

In @RasmiaKulan's project we applied the following image quality metrics to reconstructions with FBP and iterative:

Signal to noise ratio (SNR):

this requires an ROI in a dataset
calculates the mean divided by the standard deviation in the ROI
Note: we already have PSNR - do we want to have both?

Contrast to noise ratio (CNR):

this requires two different ROIs to compare the contrast between
Calculates [mean(ROI1) - mean (ROI 2) ]/ stdev ROI2
There are other methods of calculating this so which one to implement is to be discussed

L2-Norm-Error:

takes two different images or ROIs in two different images
this is an assessment of the difference in pixel values
L2Norm(image1-image2)/L2Norm(image2)

We could add these image quality metrics into CIL, where we currently have some existing metrics: https://github.com/TomographicImaging/CIL/blob/4f6e3cfb0648d08071974eba35c1369a892d4924/Wrappers/Python/cil/utilities/quality_measures.py#L4

Some of the existing metrics, such as PSNR already exist in skimage.metrics so we could potentially make use of this instead and remove them from CIL https://scikit-image.org/docs/stable/api/skimage.metrics.html

lauramurgatroyd commented 2 months ago

Also related: https://github.com/TomographicImaging/CIL/issues/1836