Detail visibility - contrast to noise ratio
19.11.2019, 11:41
As mentioned in there is a second parameter which defines the detail visibility: The contrast. While unsharpness shows the resolution perpendicular to the x-ray beam, contrast defines the resolution in direction of the beam. As far as I know there is only one standardized IQI for measurement of the contrast: ASTM E1647 with recesses large enough to exclude unsharpness issues (beside E2597 grooved stepwedge):
All other IQIs measure contrast and spatial resolution at the same time with the IQI, e.g. hole penetrameter, step-hole penetrameter or wires. To show that, an 8mm steel weld with an ISO19232-1 wire set was radiographed with the best film (C1) and with a DDA with 200µm large pixels - both with no magnification. The film was digitized with a high class film scanner and both images highpass filtered to show the differences better on a standard monitor.
The film image can show the wire 16 with 100µm (~1.25% contrast resolution); with the same visibility the DDA with 200µm pixel size can show the 50µm wire (~0.625% contrast resolution) - which is just 1/4 of the pixel size.
There is a second test to prove the sub-pixel visibility. To avoid too high contrast due to the more absorbing wire, grooves were used here with a prototype of the five groove wedge.
The detector has 400µm pixel size (PerkinElmer RID512/400 AF1) and can show the 127µm groove:
You can see the groove - but you could only measure in the range of 400µm. What happened to make the small grooves visible? Let's take an example with a -2% contrast groove which covers just 1/4 of a pixel in one direction:
When the Signal to Noise Ratio (SNR) is high enough, the groove is visible; as the groove is a very long "feature", it would be visible with a Contrast to Noise Ratio (CNR) of 1 - in the example with a SNR of 200 and more.
For evaluation of the visibility of wires (ISO19232-1) a test was done with a 2 focal spot tube (1.0 and 0.4 to EN12543) on a 10mm and 20mm steel plate with a DDA with 200µm pixel size (PerkinElmer XRD1621). The visibility of a wire required as the minimum a 10mm continious visible line; additionally it was measured to have a CNR of 1 for this size. The smallest wire has 50µm diameter. Different SNR levels were archivied with different integration times (detector calibration was done for a SNR of 2000 with a moving steelplate in the beam).
The graph show that the visibility of the wires depends mainly on the SNR. With 10mm steel the focal spot size is nearly independed, with more material thickness the smaller focal spot is better - but of course needed factor 2.5 longer exposure times.
As wires are not so popular in all regions of the world, the test was repeaded with holes. A step hole IQI (ISO19232-2) was used for the next picture with similar conditons as before. The visibility of the holes were evaluated and a hole had to have a CNR >2.5 to be rated as visible. The smallest hole has a diameter of 125µm (5/1000").
With the hole type IQIs the result is very similar - the SNR is the important parameter which determines if a hole is visible (has a stronger signal compared to the noise level).
Conclusion: If the noise level is very low, you can see details much smaller than your pixel. To be honest - you knew that before; if you ever take a picture were a powerline was in, you will see the - compared to the distance - really small wires in your picture (and sometimes have to use the Photoshop to remove them for a perfect image).
In X-ray images you should not rely on the sub-pixel resolution, because sometimes a pixel becomes a bad pixel. Please see the Fig 11 in the DDA guide ASTM E2736 for aqualitative guideline for detectability of a feature with respect to contrast-to-noise (CNR).
All other IQIs measure contrast and spatial resolution at the same time with the IQI, e.g. hole penetrameter, step-hole penetrameter or wires. To show that, an 8mm steel weld with an ISO19232-1 wire set was radiographed with the best film (C1) and with a DDA with 200µm large pixels - both with no magnification. The film was digitized with a high class film scanner and both images highpass filtered to show the differences better on a standard monitor.
The film image can show the wire 16 with 100µm (~1.25% contrast resolution); with the same visibility the DDA with 200µm pixel size can show the 50µm wire (~0.625% contrast resolution) - which is just 1/4 of the pixel size.
There is a second test to prove the sub-pixel visibility. To avoid too high contrast due to the more absorbing wire, grooves were used here with a prototype of the five groove wedge.
The detector has 400µm pixel size (PerkinElmer RID512/400 AF1) and can show the 127µm groove:
You can see the groove - but you could only measure in the range of 400µm. What happened to make the small grooves visible? Let's take an example with a -2% contrast groove which covers just 1/4 of a pixel in one direction:
When the Signal to Noise Ratio (SNR) is high enough, the groove is visible; as the groove is a very long "feature", it would be visible with a Contrast to Noise Ratio (CNR) of 1 - in the example with a SNR of 200 and more.
For evaluation of the visibility of wires (ISO19232-1) a test was done with a 2 focal spot tube (1.0 and 0.4 to EN12543) on a 10mm and 20mm steel plate with a DDA with 200µm pixel size (PerkinElmer XRD1621). The visibility of a wire required as the minimum a 10mm continious visible line; additionally it was measured to have a CNR of 1 for this size. The smallest wire has 50µm diameter. Different SNR levels were archivied with different integration times (detector calibration was done for a SNR of 2000 with a moving steelplate in the beam).
The graph show that the visibility of the wires depends mainly on the SNR. With 10mm steel the focal spot size is nearly independed, with more material thickness the smaller focal spot is better - but of course needed factor 2.5 longer exposure times.
As wires are not so popular in all regions of the world, the test was repeaded with holes. A step hole IQI (ISO19232-2) was used for the next picture with similar conditons as before. The visibility of the holes were evaluated and a hole had to have a CNR >2.5 to be rated as visible. The smallest hole has a diameter of 125µm (5/1000").
With the hole type IQIs the result is very similar - the SNR is the important parameter which determines if a hole is visible (has a stronger signal compared to the noise level).
Conclusion: If the noise level is very low, you can see details much smaller than your pixel. To be honest - you knew that before; if you ever take a picture were a powerline was in, you will see the - compared to the distance - really small wires in your picture (and sometimes have to use the Photoshop to remove them for a perfect image).
In X-ray images you should not rely on the sub-pixel resolution, because sometimes a pixel becomes a bad pixel. Please see the Fig 11 in the DDA guide ASTM E2736 for aqualitative guideline for detectability of a feature with respect to contrast-to-noise (CNR).
Last edited by admi on 17.03.2020, 20:52, edited 3 times in total.
Reason: Image Host changed
Reason: Image Host changed
Re: Detail visibility - contrast to noise ratio
17.12.2019, 21:09
In most standards "contrast visibility" is measured with hole type plaques or wires - but this ignores the influence of the size of the object to be detected and also that nearly all X-ray images are noise dominated due to the low amount of X-ray quants compared to visual camera inspection.
You may do your own test and try to clearly identify the holes in the three following images:
In the first image it is easy to identify all three holes, independend on their size.
In the second image you will "see" the smallest hole because due to the first image you know where it is. But - would you identify the same hole in another position in the image?
You may try now:
If you switched with your eyes between the upper and lower image you should admit that you did not identify this additional 1T hole.
In the third image the noise level is higher and here it is even problematic to identify the 2T hole:
The following drawing should give you an idea why the size is relevant in a noise dominated image:
(Please see for explanation the drawing with one hole only: )
For more detailed information about the visibility in noisy image please refer to the where the Rose criterium is explained.
Finally here is the solution of the task with the additional 1T hole in the second image:
and the eagle eye experts may have seen that the 1T hole is not in the center of the box, it is on 8 o'clock
You may do your own test and try to clearly identify the holes in the three following images:
In the first image it is easy to identify all three holes, independend on their size.
In the second image you will "see" the smallest hole because due to the first image you know where it is. But - would you identify the same hole in another position in the image?
You may try now:
If you switched with your eyes between the upper and lower image you should admit that you did not identify this additional 1T hole.
In the third image the noise level is higher and here it is even problematic to identify the 2T hole:
The following drawing should give you an idea why the size is relevant in a noise dominated image:
(Please see for explanation the drawing with one hole only: )
For more detailed information about the visibility in noisy image please refer to the where the Rose criterium is explained.
Finally here is the solution of the task with the additional 1T hole in the second image:
and the eagle eye experts may have seen that the 1T hole is not in the center of the box, it is on 8 o'clock
Hosted by iphpbb3.com
Impressum | Datenschutz