Bart_van_der_Wolf
pro member
Hi Folks,
Given the huge popularity of my Autofocus Microadjustment tool, I thought it might be interesting to also provide a tool to allow testing for resolution. It can be printed with any reasonable quality (inkjet) printer, and therefore offers a low cost solution for those who need to verify the resolution of their imaging chain (e.g. to test if a newly purchased lens performs well).
What is so special about it?
So what is needed is a type of target that is made from smooth, e.g. sinusoidal, cycles of brightness that have well understood characteristics and that are relatively easy to analyse, also numerically (e.g. with Fourier analysis). We can use a bar with increasingly higher spatial frequencies (see example) and observe that, as the spatial frequency increases, the recorded contrast will decrease, until we reach the absolute limiting resolution where contrast reaches zero. The drawback of such a bar test target is that we need to calibrate the image magnification (focal length and shooting distance) if we want to make a meaningful statement about the resolution in absolute terms, e.g. cycles per millimetre. It also only determines resolution in a single direction, so it would require several tests or targets at different angles to detect orientation specific issues.
To solve that, and make the readout more accurate, I had already devised a modification of another existing type of target for analog image recording, i.e. on film or video tubes, many years ago. The original target was, AFAIK, invented by the Americans Jewell and Nutting for lens resolution testing. It consisted of 72 alternating black and white tapered sectors arranged in an 8 inch circle. This allowed to know the exact distance between the segments at any given diameter. The Siemens company developed a test method based on the Jewell-star, therefore it became also known as a Siemens-star.
The drawback of sharp high spatial frequency edges in that original target makes it unsuitable for reliable testing of discrete sampling systems (such as digicams), hence my modification to a sinusoidal type of radial grating. One of the very useful properties of such a star target is that it is insensitive to differences in image magnification ratios caused by focal length and shooting distance. After all, all spatial frequencies are available at different diameters regardless of recording size. The only thing that matters is a measurement of diameter. All we need to make sure of, is that the finest details are smaller than the resolution capabilities of the best component our imaging chain.
This new test target avoids the error generating sharp edges, and records real resolution in many orientations with a single shot!
Anything smaller than the sampling density at the Nyquist frequency will be either blurred to zero contrast or aliased (also depending on things like anti-aliasing filters and/or defocus).
Aliasing artifacts will stand out by their seemingly hyperbolic divergence from the expected radial direction.
What can it be used for?
The target can be used to test printers, and/or lenses plus cameras. It can serve to determine maximum resolution of a lens in its center and/or corners, and unveil asymmetry (e.g. due to decentering, or vibration). It can help to verify the acceptable DOF limits. It can visualize the trade-off by your camera or Raw converter between detail and artifacts when demosaicing the Bayer CFA filtered data. It also allows to determine differences in resolution from the same file when processed by different Raw converters, or Noise reduction algorithms. The resolution target allows to determine an absolute number for the limiting resolution, but also gives a visual impression, especially about potential artifacts (an insight which can be useful when comparing cameras to be used for certain tasks).
Where can you find it for download?
Now, the target itself. You can download a 16-bit/channel RGB file for printing (right mouse button click for Save-as);
for HP/Canon inkjet printers (10.8 MB)
for Epson inkjet printers (15.6 MB)
Warning: People prone to epileptic response when viewing alternating bright and dark image patterns are advised to not look/stare at the pattern, especially when zoomed in.
How do you print it?
Print it at the indicated PPI without printer enhancements on glossy Photopaper. That PPI wiil usually be set by the printer driver when you select its maximum quality settings for glossy paper and set the correct size (130x130mm).
The target itself has no ICC colorspace profile embedded, so I suggest to assign the printpaper output profile to the file before you actually print it with that same profile. Doing so will probably keep the image brightnesses in the gradients distributed as evenly as intended. If you don't have a profiled print path, then just use the default printer driver's possibilities for your choice of paper and let the printer manage the translation. The step-wedge grayscales allow to recalibrate the image for numerical evaluation after regular gamma adjustment if necessary, but they also serve as a visual guide to evaluate neutral print quality.
How do you use it?
This should produce a 130x130mm test target, that can reveal issues with your printer (e.g. irregular paper feeding, or too much ink). You should shoot it with your (digi)cam from a (non-critical) distance like between 25-50x the focal length. Most good (inkjet) prints on glossy paper can achieve something like a 0.1 mm (254 PPI) resolution or better (despite some ink bleed), so at 25x the focal length distance the optical resolution should be better than most regular lenses can resolve in air, and it certainly outresolves most (if not all) sensor arrays by the time we reach 50x focal length distance.
For tests of performance at very large shooting distances, the sensor image may become too small for practical evaluation. In that case you can print the image larger (thus at a lower PPI), which should also tell you something about you enlargement algorithm's quality. Again, the shooting distance is not critical, just experiment and make sure you keep enough distance to get some low contrast blur in the center of the resulting image. The diameter of the blur center will only be affected by limiting resolution, not the shooting distance!.
The diameter of the resulting "blur"center is a measure of "on-sensor resolution" of the whole optical chain (lens + AA-filter + sensor), and can be expressed as cy/mm after calculating "(144 / pi) / diameter". The diameter can be expressed as number of pixels multiplied by the pixel pitch. So, for an example, (144 / pi) / (100 pixels x 0.0064 millimetre) = 71.6 cycles/mm on a 6.4 micron sensel pitch camera. One can also express it as 0.46 cycles/pixel, which would be close to the maximum reliable resolution, the Nyquist frequency at 0.5 cycles/pixel.
If you want to compare to different sized sensor arrays, all you need to do is scale it to the difference in physical sensor size. After all, physically larger sensor arrays, require less output magnification to reach a given output size.
For digital images one can use e.g. the ruler tool in Photoshop to measure the number of pixels in the central blur diameter at various orientation angles.
For print or analog film evaluation you can use an optical microscope and a reticle, or if you scan your film you can use a scanner to evaluate the entire imaging chain (camera lens + film + scanner).
Using this target may reveal some shortcomings in your equipment or workflow, so you are warned. Do not blame me for allowing you to detect it.
Closing remarks.
I've added some other features to the test target, such as slanted edges for MTF determination and a simple detector for wrong gamma (which yet may require some tweaking). I also added a copyright notice to warn against unauthorized reproduction of the target, but obviously you can download and print the target for your personal use. Feel free to ask questions if anything is not clear.
Cheers,
Bart
P.S. I've added Red/Blue versions of the above targets for testing the worst (but unlikely) case scenario for a Bayer CFA:
for HP/Canon inkjet printers (14.3 MB)
for Epson inkjet printers (19.6 MB)
P.P.S. I've added a few patches that can be used to quickly verify the print quality of the target. This is in reaction to recent feedback where a sub-optimal print quality affected the capabilities to get good lens test results. The new features are explained here.
Given the huge popularity of my Autofocus Microadjustment tool, I thought it might be interesting to also provide a tool to allow testing for resolution. It can be printed with any reasonable quality (inkjet) printer, and therefore offers a low cost solution for those who need to verify the resolution of their imaging chain (e.g. to test if a newly purchased lens performs well).
What is so special about it?
- It's cheap to produce yourself, it's virtually free!
- It is accurate, and it produces repeatable results!
- It's easy to use, no need for exact shooting distances!
- It clearly reveals a camera's tendency for aliasing!
- It allows to compare Raw converter capabilities (resolution versus artifacts)
So what is needed is a type of target that is made from smooth, e.g. sinusoidal, cycles of brightness that have well understood characteristics and that are relatively easy to analyse, also numerically (e.g. with Fourier analysis). We can use a bar with increasingly higher spatial frequencies (see example) and observe that, as the spatial frequency increases, the recorded contrast will decrease, until we reach the absolute limiting resolution where contrast reaches zero. The drawback of such a bar test target is that we need to calibrate the image magnification (focal length and shooting distance) if we want to make a meaningful statement about the resolution in absolute terms, e.g. cycles per millimetre. It also only determines resolution in a single direction, so it would require several tests or targets at different angles to detect orientation specific issues.
To solve that, and make the readout more accurate, I had already devised a modification of another existing type of target for analog image recording, i.e. on film or video tubes, many years ago. The original target was, AFAIK, invented by the Americans Jewell and Nutting for lens resolution testing. It consisted of 72 alternating black and white tapered sectors arranged in an 8 inch circle. This allowed to know the exact distance between the segments at any given diameter. The Siemens company developed a test method based on the Jewell-star, therefore it became also known as a Siemens-star.
The drawback of sharp high spatial frequency edges in that original target makes it unsuitable for reliable testing of discrete sampling systems (such as digicams), hence my modification to a sinusoidal type of radial grating. One of the very useful properties of such a star target is that it is insensitive to differences in image magnification ratios caused by focal length and shooting distance. After all, all spatial frequencies are available at different diameters regardless of recording size. The only thing that matters is a measurement of diameter. All we need to make sure of, is that the finest details are smaller than the resolution capabilities of the best component our imaging chain.
This new test target avoids the error generating sharp edges, and records real resolution in many orientations with a single shot!
Anything smaller than the sampling density at the Nyquist frequency will be either blurred to zero contrast or aliased (also depending on things like anti-aliasing filters and/or defocus).
Aliasing artifacts will stand out by their seemingly hyperbolic divergence from the expected radial direction.
What can it be used for?
The target can be used to test printers, and/or lenses plus cameras. It can serve to determine maximum resolution of a lens in its center and/or corners, and unveil asymmetry (e.g. due to decentering, or vibration). It can help to verify the acceptable DOF limits. It can visualize the trade-off by your camera or Raw converter between detail and artifacts when demosaicing the Bayer CFA filtered data. It also allows to determine differences in resolution from the same file when processed by different Raw converters, or Noise reduction algorithms. The resolution target allows to determine an absolute number for the limiting resolution, but also gives a visual impression, especially about potential artifacts (an insight which can be useful when comparing cameras to be used for certain tasks).
Where can you find it for download?
Now, the target itself. You can download a 16-bit/channel RGB file for printing (right mouse button click for Save-as);
for HP/Canon inkjet printers (10.8 MB)
for Epson inkjet printers (15.6 MB)
Warning: People prone to epileptic response when viewing alternating bright and dark image patterns are advised to not look/stare at the pattern, especially when zoomed in.
How do you print it?
Print it at the indicated PPI without printer enhancements on glossy Photopaper. That PPI wiil usually be set by the printer driver when you select its maximum quality settings for glossy paper and set the correct size (130x130mm).
The target itself has no ICC colorspace profile embedded, so I suggest to assign the printpaper output profile to the file before you actually print it with that same profile. Doing so will probably keep the image brightnesses in the gradients distributed as evenly as intended. If you don't have a profiled print path, then just use the default printer driver's possibilities for your choice of paper and let the printer manage the translation. The step-wedge grayscales allow to recalibrate the image for numerical evaluation after regular gamma adjustment if necessary, but they also serve as a visual guide to evaluate neutral print quality.
How do you use it?
This should produce a 130x130mm test target, that can reveal issues with your printer (e.g. irregular paper feeding, or too much ink). You should shoot it with your (digi)cam from a (non-critical) distance like between 25-50x the focal length. Most good (inkjet) prints on glossy paper can achieve something like a 0.1 mm (254 PPI) resolution or better (despite some ink bleed), so at 25x the focal length distance the optical resolution should be better than most regular lenses can resolve in air, and it certainly outresolves most (if not all) sensor arrays by the time we reach 50x focal length distance.
For tests of performance at very large shooting distances, the sensor image may become too small for practical evaluation. In that case you can print the image larger (thus at a lower PPI), which should also tell you something about you enlargement algorithm's quality. Again, the shooting distance is not critical, just experiment and make sure you keep enough distance to get some low contrast blur in the center of the resulting image. The diameter of the blur center will only be affected by limiting resolution, not the shooting distance!.
The diameter of the resulting "blur"center is a measure of "on-sensor resolution" of the whole optical chain (lens + AA-filter + sensor), and can be expressed as cy/mm after calculating "(144 / pi) / diameter". The diameter can be expressed as number of pixels multiplied by the pixel pitch. So, for an example, (144 / pi) / (100 pixels x 0.0064 millimetre) = 71.6 cycles/mm on a 6.4 micron sensel pitch camera. One can also express it as 0.46 cycles/pixel, which would be close to the maximum reliable resolution, the Nyquist frequency at 0.5 cycles/pixel.
If you want to compare to different sized sensor arrays, all you need to do is scale it to the difference in physical sensor size. After all, physically larger sensor arrays, require less output magnification to reach a given output size.
For digital images one can use e.g. the ruler tool in Photoshop to measure the number of pixels in the central blur diameter at various orientation angles.
For print or analog film evaluation you can use an optical microscope and a reticle, or if you scan your film you can use a scanner to evaluate the entire imaging chain (camera lens + film + scanner).
Using this target may reveal some shortcomings in your equipment or workflow, so you are warned. Do not blame me for allowing you to detect it.
Closing remarks.
I've added some other features to the test target, such as slanted edges for MTF determination and a simple detector for wrong gamma (which yet may require some tweaking). I also added a copyright notice to warn against unauthorized reproduction of the target, but obviously you can download and print the target for your personal use. Feel free to ask questions if anything is not clear.
Cheers,
Bart
P.S. I've added Red/Blue versions of the above targets for testing the worst (but unlikely) case scenario for a Bayer CFA:
for HP/Canon inkjet printers (14.3 MB)
for Epson inkjet printers (19.6 MB)
P.P.S. I've added a few patches that can be used to quickly verify the print quality of the target. This is in reaction to recent feedback where a sub-optimal print quality affected the capabilities to get good lens test results. The new features are explained here.
Last edited:
