custom target size and consequences of scaling the target down

[Theresa Senti]

Hi there,
I want to use the MTF targets to evaluate camera parameters for a computer vision setup. The Field of View is quite small. Since I want to evaluate the whole setup as-is, I need a target that is small enough to fit into my field of view. In this case it means, that the size of the target is more like A5 or even a bit smaller. I mainly care for MTF50 (lensgrid chart) at the moment, but the focus will be relevant as well. Now I have the following questions:
1. Is there a way to export the targets in a custom size (in the user guide I only saw sizes down to A4 but no smaller or custom options). If yes: how?
2. If it’s not possible to get a custom MTF-chart, what are the consequences of downscaling the A4 to the size I need?
3. If I have to scale it down manually, is it better to scale it in a way I can see the whole target (with the consequence that the block edges will be shorter than 25 pixels) or crop it to have bigger blocks but obviously cut away parts of the target?

Thank you in advance for your answers and your help!

[Frans van den Bergh]

Hi Theresa,

1. Sorry, but I don't currently have built-in support for generating charts smaller than A4. Note that this only applies to the code that generates the charts.
2. MTF Mapper will accept and process almost any test chart that contains black squares (or trapezoids) on a white background. The lensgrid style chart adds two things:
   a) the edges of the squares/trapezoids are reasonably well aligned with the tangential and radial directions, while avoiding a few known bad edge orientations (e.g., 0 degrees or 45 degrees). But MTF Mapper should flag edges with bad orientations with red or yellow in the annotated image output, so it is not hard to design your own chart layout if you wish.
   b) the circular fiducials are coded to reflect the page size, but these are only used if you request the "chart orientation" output. Since you can readily achieve a good chart orientation / alignment with other methods, e.g., using a small mirror, the loss of the automated chart orientation feature is not really a significant problem.
3. You can really use a combination of scaling and cropping as desired, but as you noted, you should try to keep the edges (measured in pixels in the resulting image) longer if possible. The 25-pixel minimum is really the absolute minimum, and I would recommend that you aim for > 50 pixels if your sensor resolution allows. Conversely, you want the edges to be small relative to the field of view, so I would recommend keeping the edge length below 8% of the field of view (otherwise a single edge will be a mixture of SFRs from a larger section of the lens).

Please keep in mind that print quality is more likely to become a significant issue with smaller test charts. I would recommend that you aim for 8 printer dots under every sensor pixel, although you can probably get good results with only 4-5 printer dots under every sensor pixel. If your field of view is much smaller than A5, then you might have to consider LVT transmissive targets. The best way I check whether printing quality is good enough is to compare the SFR of to edges that are close to each other, but have different print orientations (but similar lens radial/tangential orientations); if the SFRs differ a lot, then you could have a target quality issue.

I'd be happy to give you feedback on your scaled down chart if you like; just send it (chart or sample images) to fvdbergh@gmail.com, or post here if appropriate.

Regards,
Frans

[Theresa Senti]

Thank you so much for your reply Frans! I will reach out directly as soon as I have a scaled down version.