Mavic Air vignette test

[Prismatic]

Summary:
It's often difficult to characterize the effect of vignetting in normal photographs; sometimes it's apparent, other times not so much. Outdoor movie screens offer a big, uniform target for testing. Here's what I learned:

• JPG photos from the Mavic Air show quite uniform illumination across the field, but critical users may find a slight warm color cast in the central region and a vanishingly small residual vignette.
• DNG (digital negative, a 'raw' format) photos need considerable corrections to obtain uniform illumination and coloration; I have not worked out exactly what parameters would accomplish that.

For my purposes, and with the tools I use, using DNG is out of the question. Until I have a proper camera profile for the MA, one that corrects DNG images as well or better than what's built-into the MA, I'll be shooting JPGs.

NOTE: This analysis is strictly about field illumination, and only for still photos. A quick check suggested similar results with 4K video, but I didn't run a conclusive set of tests.
I also don't touch on the Mavic Air's well-known barrel distortion. Other threads discuss that.
======================================
Test setup: I live near a twin-screen drive-in theater. The blank screens provide flat, evenly illuminated "gray cards" for testing sensor illumination. At about 2PM, one screen has the sun behind it (it's opaque), the other faces the sun. Both appear uniformly white to the naked eye, excepting the small visible defects.

The four images below are from one shot: A full-resolution, full-color DNG and it's corresponding & simultaneous JPG, plus a specially processed version of each. This shot is representative of the twenty or so bracketed exposures I ran against both screens. There was no ND filter in this case, but in other shots I used a PolarPro ND8, and found no discernible difference. This image happens to be of the "shady" screen; results on the sunlit screen were identical.
======================================
This is a snapshot of the unadulterated, non-processed DNG. Strong vignetting is obvious. Perhaps more distressing--because it's harder to correct--the vignette includes a serious color-shift: the center has a distinct red/orange tone, while the corners and edges trend toward cyan (e.g., a central pixel R165, G159, B153; a corner pixel R111, G117, B119):


Below is the corresponding, simultaneous JPG. The vignetting appears to be fully corrected, and the corners and edges are perfectly neutral in color. However, while difficult to detect visually, the central area still shows a very slight extra orange/yellow warmth (e.g., central pixel R139, B137, G134; corner pixel R138, B138, G138):

======================================
I then ran histogram equalization against them both. This increases the global contrast of an image where the brightness values are nearly uniform. I used it to exaggerate the field illumination. This helps define the extent of the vignetting, and reveal the degree to which on-board image processing compensates for the in-camera vignetting. (Histogram equalization takes coloration out of consideration.)

First, here's the equalized DNG image. Wow! It's not just the extreme corners & edges, the whole image is one big vignette! It's wildly exaggerated of course, but it drives home that a DNG source will probably not produce an acceptable result without application of a suitable corrective profile:


Finally, the equalized JPG shown here reveals a trace of residual vignetting, but not very much. The previously nearly invisible rippling and tiling of the surface is now strikingly obvious. This means that in practical terms, the remaining vignetting is of little consequence, no more that could be expected in any extremely wide-angle lens:


I know of no source for a Mavic Air camera profile at present, and I myself quickly tire of fiddling with sliders! Shooting DNG, it appears, is not my cup of tea, at least not for now.

Let's hope a good corrective profile becomes available soon! Because the MA not only color-vignettes (well-corrected in JPG), there's that unfortunate barrel distortion, too, which is apparently not well-corrected in JPG, from what I see in threads. %^(

I'll run my own distortion test later this week, but expect to ride in the barrel-distortion rodeo like most. Regardless, I still really like this bird!

 

[geigy]

Very interestingi i can't recall anyone posting their results in such a complete and clear way ever! Kudos to you.

Have you considered shooting a flat field image using a relatively close up target such as a sheet of paper?

I've used photoshop's divide layer mode to flatten the field in microscopy. Very effective. I wonder if this principle could be applied until a LR camera profile becomes available..

 

[geigy]

The GIMP also has divide mode, and had it before Photoshop did. I wonder if there could be a procedure to remove vignetting that way.

 

[Prismatic]

... Have you considered shooting a flat field image using a relatively close up target such as a sheet of paper? I've used Photoshop's divide layer mode to flatten the field in microscopy. ...

Thanks for the kind words. Ages ago, I used a 4x5 wood camera, and learned then the importance of creating tests with the fewest possible variances from in-the-field photography. Actually achieving uniform illumination is way more difficult than it might seem, and small targets exponentially compound all issues.
As it happens, I'm too miserly to use Photoshop; I use a competing tool, but not GIMP (a different thread for a different day).

Apropos of nothing, I'm a photographer, astronomer, engineer, and writer. Not to mention windy.

So I can't help mentioning that, technically, a "flat field" refers to the image plane of a lens. A superbly flat-field lens focuses all stars--ideal infinitely distant point-sources of light--onto a flat sensor, needle sharp from its center to all its corners. With lesser lenses, the stars focus on a curved surface, more like the inside of a "bubble" behind the lens (like your retina!). The optical concept is called "curvature of field", and less of it the better!

My test was of somewhat different concept: field illumination. A lens that can project a uniformly gray target--say, an outdoor movie screen--as a uniformly gray flat image has "flat" field illumination. Unfortunately, wide-angle lenses--like the one aboard the Mavic Air--are prone to poor performance in this regard. Their images often have a noticeably brighter center compared to their edges and corners. Basically the lens, strongly curved by necessity, is closest to the sensor at its center. The image is brightest at that point, and fades radially, as my test illustrates. That's called vignetting, and the less of it, the better, too.

But although flat field illumination is desirable, the problem can largely be digitally corrected, as my test shows. Because of this, modern lens designers have more freedom to allow vignetting in favor of optimizing lens attributes that can't be digitally corrected--like curvature of field!

 

[geigy]

Thanks - yes, the "flat field" I was referring to should have been "flat field illumination"..

Your article inspired me to check out my Mavic Air today and it seems to be uniformly sharp, with significant burring in the corners, but good enough for video work. Good as they are, these tiny cameras don't compare with the pro gear.

Regarding curvature of field, I wonder If we will live to see an integral lens/sensor with a curved sensor. This would eliminate a couple of air/glass interfaces, and also allow extraordinary wide field of view with no cosine effect. The pixels would be non-rectangular, but, as you say, fix it in digital..

There have been some products, surprisingly, that do this a little bit with film. One was a < 1.0 lens that used a layer of oil between the rear element and the film, and another was the Minox, which had a curved pressure plate.

Once again, I enjoyed reading your article and hope to see more. I clicked on "Follow" for your posts so I don't miss them. This is a first for me on this board. All the best.

 

[Prismatic]

Thanks - yes, the "flat field" I was referring to should have been "flat field illumination"..

Your article inspired me to check out my Mavic Air today and it seems to be uniformly sharp, with significant burring in the corners, but good enough for video work. Good as they are, these tiny cameras don't compare with the pro gear.

Regarding curvature of field, I wonder If we will live to see an integral lens/sensor with a curved sensor. This would eliminate a couple of air/glass interfaces, and also allow extraordinary wide field of view with no cosine effect. The pixels would be non-rectangular, but, as you say, fix it in digital..

There have been some products, surprisingly, that do this a little bit with film. One was a < 1.0 lens that used a layer of oil between the rear element and the film, and another was the Minox, which had a curved pressure plate.

Once again, I enjoyed reading your article and hope to see more. I clicked on "Follow" for your posts so I don't miss them. This is a first for me on this board. All the best.

“... significant burring ...”? “Blurring” was your intent, I reckon, and I’m not too surprised. Optical design at this scale is an exercise in massive compromises; what’s amazing is that these cameras perform at all, defects aside! I’m more than happy.
Interesting ideas there, especially about non-rectangular pixels. And the Minox! Loved them, real 007 stuff! (Terrible grainy image though.)
Advances in lens-free imaging @Caltech, @Hitachi, and elsewhere may render classical optics obsolete soon anyway, and that will change everything. It’s likely that technology will break out before curved sensors become needed for increased performance in micro cameras.
Meanwhile, an amazing saga in technoforensics has been playing out in the current “Mavic Air flyaway” thread here. If you haven’t been following it, start at the top, don’t jump to the end. It’s a great, great read!