When I bought the X70 a year ago, I chose it over the Ricoh GR II (basically the only other camera in the same class, APS-C premium compacts).

I made my decision then primarily based on specifications; the X70 was the newer model, has phase-detect autofocus (which is theoretically quicker than the GR’s contrast only AF) a tilt and touch-enabled LCD, and better video capabilities. The X70 further distinguishes itself from the GR by having absolute value physical controls for aperture, shutter speed, and exposure compensation (but not for flash).

Both cameras are aimed at a specific kind of photography use case: that of the high quality pocketable camera; a camera for street photographers, its small size and discreet appearance of making it more likely that you’ll have the camera with you to capture spontaneous, intimate scenes which could otherwise be missed.

The most most noticeable difference between the two cameras is probably the size and weight: the X70 is about 92 grams heavier than the GR, and slightly larger (just large enough to keep it from fitting in a jeans pocket). Both cameras have similar image quality, the Ricoh being slightly superior in both optics and sensor.

This is not a intended to be a complete review or comparison, just a re-cap of my experience in using the X70 in the real world for a year.

Macro (Close up)

One area where the X70 excels is in shooting close ups. The minimum focus distance is 10cm, there is no need to switch into a special “macro” focus mode, and the built in flash illuminates the subject fairly evenly at this distance. The lens is also sharp close-up and wide-open (unlike many other Fujinon lenses).

General Street/Documentary

Although the X70 is fairly discrete, it is not really pocketable. No more than say an X100T anyway (which is only another 93 grams heavier and has a viewfinder). It’s heavy enough that carrying it in a coat pocket requires a counterweight in the opposite pocket to maintain the balance of the garment, and you’re definitely not going to fit it in a pants pocket (unless they’re cargo pants). If you do put it in a jacket pocket, or a bag, or anything else, expect the settings to be different upon removing it, as the dials are quite easily disturbed by handling.

While the X70’s tilt screen is an asset in theory and sets it apart from the Ricoh GR, in practice I found that it was it was too slow to deploy for use on the street. Most of the time, in the interest of speed, I left it flat and squatted or shot blindly from the hip just as I would have done with a fixed-screen or a camera having a viewfinder. It did occasionally come in handy — just not while shooting street. I will say that the tilt screen is robustly made — it’s actually quite a bit sturdier feeling than the one on the newer and much more expensive X-T2.

The lens cap situation is another speed bump. The GR has an integrated leaf-style lens cover that opens and closes automatically. The X70, however, comes with a 1970s style metal and felt slip on lens cap, which is just too fiddly to use. Instead, one must buy the lens hood/filter adapter accessory, which permits both the use of a pinch-style cap and (my preference) a protective clear glass filter (which has the benefit of preventing the camera from ingesting dust through the lens).

One absolute physical control the X70 has and the Ricoh GR lacks that I do find useful is the power switch — the X70 has the power switch around the shutter button, like the shutter lock switch found on many film cameras. However, like all of the other dials, this one tended to move on its own, so I would sometimes remove the camera from my bag and find the switch already in the “on” position, but the camera “asleep.” This switch needs to be stiffer, which seems to be a common theme with the X-Series.

The physical control that is conspicuously absent is that for enabling the flash; in order to turn the flash on and off, you have to look at the screen and enter a series of button presses into the back of the camera. This makes it difficult to adapt to the rapidly changing conditions of street shooting.

Even though it’s not truly pocketable, it is a small camera, and, as such, is less intimidating to subjects. This can allow one to get very close indeed without startling people — unless you look like me, anyway, in which case you’ll still get reactions like this:

The results are a little better when the subject is too absorbed in their activity to make eye contact…

Nature/Landscape

Due to its size and weight and having a tilt screen, I thought the X70 would be a great camera to take hiking. I thought wrong. I took it as my only camera on the Silver Falls Trail of Ten Falls loop and took a plenty of pictures, but none of them turned out to my satisfaction. This is partly due to the 18mm (28mm equiv.) focal length. It isn’t quite wide enough to capture a waterfall close up and isn’t long enough to get one from a distance. Perhaps the WCL-X70 adapter (21mm equiv) would have made a difference here. However, the X-Trans sensor struggles to render detail in foliage and rock, and I think with a wider angle optical adapter the results would have still been disappointing.

I took it to the Oregon coast and found that the X70 came up short there too (although on this trip it wasn’t my only camera and I did get some good shots with the X-Pro1 and 14mm F2.8).

In summary, I’ve never got a landscape shot out of this camera that I liked, and it’s not for want of trying.

For a little more money (depending on sale prices) than the the X70 + WCL-X70 (533 grams combined), one could get an X-Pro1 and 14mm F2.8 (689 grams combined) and get better results.

World Naked Bike Ride (2016, Portland, Oregon)

The WNBR turned out to be a very interesting use case for this camera and the one in which it performed its best.

I had a severe case of battery anxiety, so I brought a spare battery for the X70 and a USB power bank, and used an external flash powered by two AAA batteries (with spares for those too). The flash was the FujiFilm EF-X20, which has its own quirks…

The EF-X20 turns turns off whenever the camera is turned off or goes to sleep, or after a timeout period of 15 minutes. And, unlike other flashes which go to sleep, the EF-X20 never wakes back up on its own. The flash turning off when the camera goes to sleep can be prevented by taping over the TTL pins (in which case the flash can only be used in manual power mode), however the timeout cannot be disabled. I had to regularly check the if the flash was on and, if it had turned off, and turn it back on — while riding. The only positive thing I have to say about the EF-X20 is that it’s very small and it has manual power control, and I say that through gritted teeth.

I wore the camera on a strap around my neck. I disabled the touch screen function, as contact the bare skin of my chest would otherwise have moved the focus point around.

I used AF-S in the central zone mode and a relatively wide aperture (for this kind of thing) of around f/4-f/5.6 (the aperture tended to move around on its own as I handled the camera) to admit more ambient light and reduce the flash power requirement. I shot at ISO 800 and with a slow shutter speed for the same reasons. The manual power setting of the EF-X20 varied from 1/16th to 1/64th, depending on subject distance (and of course this setting also moved around unbidden).

It turned out that I didn’t need all those batteries — I ended the night only 30% into the first one. I’m not a spray and pray type though, and I only shot 118 frames, 15 of which (12.7%) were keepers. Most of the missed shots were due to AF failures and flash failures (flash had turned itself off or the power dial had moved by itself). Even so, 12.7% is a pretty good keeper rate for street shooting — especially one handed while riding a bicycle naked.

The AF worked OK for subjects at a fixed distance (it took a few seconds to lock on), but failed miserably for subjects at a varying distance (i.e. approaching or receding). The shutter lag was difficult to adapt to due to its variable duration.

Many people remarked on the camera and found it stylish. I think it was most successful as a fashion accessory (however, the gold fanny-pack received just as much, if not more, attention).

Conclusions

Although I have written several articles discussing the shortcomings of FujiFilm’s X-Trans technology, the issues of usability and ergonomics are of far greater practical importance, especially in consideration of the demanding constraints of street photography.

It is because of these deficiencies that I have decided to part with the X70. Many of the ergonomic and usability problems with the X70 are common to all X-Series cameras and seem to be a core part of FujiFilm’s design language. It seems highly unlikely that FujiFilm would do away with their absolute value physical dials, for instance. With the other cameras in the X-Series, this design presents less of a problem. But with the X70’s ostensible pocketability, the easily bumped absolute value controls are more hindrance than help. The camera is just heavy and large enough to keep it from being truly pocketable, which was the main thing I wanted it to be.

The upcoming replacement model (X80?) may improve the size and weight situation, but all of the other problems are likely to remain unaddressed.

Overall, while I have enjoyed the images I made with it, I never reached a point where using the X70 was comfortable and convenient. It was always a bit of a hassle, requiring constant vigilance with the settings, fiddling with lens cap, navigating the flash menu, etc.

If you want to carry it on a strap and have time and free hands enough to adjust the settings, this could be a good camera for you (although with those constraints you might as well go for a model with a viewfinder).

If you want to carry it in a pocket and be able to draw quick and shoot from the hip, then you might want to keep looking.

It is certainly a stylish camera, but, in my opinion, it is a case of placing form over function. It’s clear that FujiFilm doesn’t really understand the needs of photographers in this market segment. Performance in the lab or the showroom floor is an entirely different thing from performance on the streets and specs don’t matter really matter if you don’t get the shot.

Jonathan Moore Liles is a photographer, writer, musician, and software architect living in Portland, Oregon.

• @nevermindhim * Instagram photos and videos
• Gallery | Nevermind Him - The Photography of Jonathan Moore Liles

Since the introduction of the FujiFilm X-Series of cameras, reviewers and consumers have struggled to compare them directly to the competition. FujiFilm’s is a tightly integrated system, wherein everything is a little bit different. They rate ISO by a different standard, use a non-standard Color Filter Array, and their RAW files rely on proprietary metadata to correct exposure levels (which 3rd party RAW processors may ignore). One well known lens and camera metrics concern has declined to even attempt any comparison against FujiFilm X Series cameras.

The effect of all these confounding factors, intentional or not, along with FujiFilm’s hyperbolic and cryptic marketing copy, has been to lead consumers to draw incorrect conclusions when comparing FujiFilm against other brands of camera, specifically regarding noise, moiré, and detail resolution. If you compare a FuijFilm camera to another brand of camera without accounting for these various factors, you may think the FujiFilm performs better in every regard. You may even think that there’s something magical about it.

Given this confused situation, I wanted to perform a comparison which eliminates all of these factors and compares the Bayer CFA to the X-Trans CFA as directly as possible, without involving lighting, optical aberrations, lens light transmission, ISO ratings, noise reduction, optical lowpass filters, etc., and see if the X-Trans CFA really does offer any of the advantages FujiFilm claims it does when compared against Bayer — with no marketing funny-business.

As the saying goes, the devil’s in the details, and there are a lot of details involved, so grab your spectacles and wand: we’re going in search of that Fuji X Magic!

Methodology

In order to remove the complicating factors of optics, base sensor technology, etc., these comparisons are performed with synthesized raw images. This is what raw data from a sensor with a Bayer and an X-Trans CFA looks like, respectively, before demosaicking:

This methodology allows for a direct comparison of the output with the input images (ground truth), and is the same technique employed by the researchers who develop demosaicking algorithms. The goal is to simulate an AA-filterless sensor, differing only in CFA (Bayer vs. X-Trans). The synthetic raw images are generated by filtering the target images through the respective CFA patterns. The resulting data is then fed into a demosaicking algorithm. DCRaw is used for all demosaicking because it conveniently allows us to provide our own raw pixel data without having to wrap it in a container. The target images themselves have been downscaled significantly from their original size in order to eliminate any noise and false colors from the input.

Because FujiFilm’s own X-Trans demosaicking algorithm is proprietary, it could not be used for this comparison. Instead, I use Frank Markesteijn’s algorithm (in highest quality 3-pass mode). However, as I have shown in my previous articles, Eking the Most out of X-Trans: Can Free Software Beat Iridient?, and X-Trans: The Promise and the Problem, this algorithm is at least as good as (and perhaps better than) FujiFilm’s.

For the Bayer images, we use AHD, a similar high quality algorithm for demosaicking Bayer, which shares some properties with the Markesteijn algorithm. There are better algorithms available for Bayer, but this is the best DCraw supports. (I prefer AMaZE, myself.)

Performance

FujiFilm representatives have tossed around various figures of their own for X-Trans performance (“30% slower”) and have also hinted that one of the reasons they choose Bayer for their new medium format GFX 50S camera was because demosaicking 50 megapixel X-Trans images would be too slow.

In these tests, X-Trans demosaicking took approximately 3.27 times as long as Bayer. So perhaps what FujiFilm really meant to say is that X-Trans demosaicking is 30% as efficient as Bayer. In any case, it’s significantly slower, which fact is uncontested.

Bayer vs. X-Trans

Those who have been following this series of articles will be familiar with this image — possibly even sick of it. The difference between these two patterns of red, green, and blue sensitivity is what this article is all about.

The Bayer CFA is common and very well established. Invented by Bryce Bayer of Kodak in the early 1970s, the Bayer CFA has been part of digital photography since its inception. FujiFilm introduced X-Trans promising that it offered many improvements over Bayer, most of them incredible.

FujiFilm claims of X-Trans that:

“The unique random color filter array reduces moiré and false colors without an optical low-pass filter. These color filters also have the effect of increasing the resolution so, when shooting with a high-resolution Fujinon lens, the camera delivers images with a perceived resolution far greater than the actual number of pixels used.”

As I’ve pointed out in previous articles, and as you can plainly see for yourself in the above figure, there’s absolutely nothing random about the X-Trans CFA. It’s just a larger pattern, 6x6 vs Bayer’s 2x2. To call it random is extremely misleading, but that seems to be the theme for the entire brochure.

Here FujiFilm elaborates on the claim that APS-C X-Trans can match the performance of full-frame (presumably higher resolution as stated above) Bayer:

“The FUJIFILM X-M1 is equipped with a large APS-C X-Trans CMOS Sensor, which offers picture quality comparable to that of full-frame sensors. The sensor’s unique colour filter array minimises moiré and chromatic aberration without the need for an optical low pass filter, while dramatically boosting resolving power even at identical pixel counts to deliver sharp and texture-rich pictures.”

FujiFilm seems to be conflating false color (a type of aliasing) and chromatic aberration (a property of lenses) here, but it’s the dramatically boosted resolving power that’s fantastic. All of these are rather bold claims, to say the least, which have never, to my knowledge, been backed up by any evidence. But that’s OK. We can test these claims ourselves — cast a spell of knowledge, if you will…

Moiré and False Color

Let’s begin with a standard test-chart like image designed to show the limits of resolution.

Well, the moiré/false color certainly looks different. Whether it’s reduced or not appears to depend on the hue of the subject.

I think we can do better than charts though, let’s try a subject where one typically encounters moiré in the real world: fabric. (Note that the input image is completely monochrome.)

Well, so much for that… This example looks far worse with X-Trans! The fact is, some patterns and orientations will look better with Bayer and some will look better with X-Trans. This is simply because the CFA patterns are different and will therefore interfere with different subject patterns. Just different, still patterns — neither are random. I see no evidence that is one inherently more resistant to moiré than the other.

So how does FujiFilm deal with this? How have their claims of moiré reduction gone unchallenged? Well, as we discovered in the first article of this series, the in-camera processing does more than just demosaic the image, it also applies heavy chroma noise reduction and color profiles which reduce overall saturation (these are called Film Simulations by FujiFilm). Let’s simulate this effect by applying strong bilateral filtering to the chroma and reducing the saturation a bit (by about as much as STD/Provia does).

Well, that’s more like it. Now the moiré is significantly attenuated. However, there’s a problem: we could have done just the same kind of filtering with Bayer (or any other CFA)! Using X-Trans didn’t buy us anything. But that’s not the only problem with this approach. Let’s see what happens when we apply the same processing to an image with color:

Ah, there’s the rub. This spell has a catch! The butterfly wings didn’t fare too poorly (the high contrast edges make it easy for the bilateral filter), but look at the unnatural color of the fingernail in the NR’d image! This happens because the chroma NR strength required to eliminate moiré, which is, in-camera, naively applied to all images, regardless of the actual presence of moiré, is much greater than the chroma NR strength required to eliminate chroma noise. By using a technique designed to treat color noise to treat both color noise and false color (which have similar appearance but different causes), fine and especially subtle color variations are lost, even in low noise, low ISO images. FujiFilm doesn’t mention it anywhere in their marketing copy, but this is how their X-Trans cameras suppress moiré. Not optically, not in the particular arrangement of the CFA, but purely in the digital domain, algorithmically, and in a general way that is equally applicable to AA-filterless Bayer images.

If you think the effect on a fingernail is bad, take a look at was this kind of processing does to a face:

Observe the color of the teeth and eyes, and how the skin has taken on a waxen, lifeless appearance. You can pump the saturation back up all you want, but you can never recover the fine color detail after this kind of processing has been performed.

Resolution and Fidelity

For each example we present the ground truth, followed by the Bayer (left) and the X-Trans (right) results, below these we show the respective difference images (that is, the difference from the ground truth).

Example 1

Example 2

Example 3

Example 4

False colors are most apparent with high contrast monochrome subjects, which are instructional if not realistic. (Image courtesy WikiMedia Commons.)

PSNR Stats

PSNR is a standard measurement for quantifying image degradation. In this case it measures the difference between the ground truth and the demosaicked output. The bigger the number, the higher the fidelity.

Example       Bayer       X-Trans       Winner
1             31.36       30.62         Bayer
2             29.30       29.38         X-Trans
3             35.13       35.04         Bayer
4             23.43       22.94         Bayer

Bayer wins overall. From looking at the difference image, it seems likely that if AHD were doing a better job interpolating the near-diagonal lines in Example 2, it would have won across the board. Notably, X-Trans performed poorly on Example 1, which contained a lot of red. This is because there are fewer red and blue sites in the X-Trans pattern compared to Bayer. The poor performance of X-Trans on Example 4 is more interesting. With a monochrome subject, the extra green in the X-Trans pattern is supposed to produce a perceived higher fidelity result than Bayer. The reality is that the false colors wash out this supposed advantage and it actually performs worse. There is no evidence of “dramatically boosted resolving power.”

What about noise?

The preceding tests were conducted with images containing essentially no noise. It has been stated by reviewers that the X-Trans CFA offers a noise advantage over Bayer, producing lower color noise and having a more “film like” grain. Let’s see if that’s true. In order to test this, we generate a noise image and apply the very same noise image to the raw Bayer and X-Trans data before demosaicking, which simulates how noise occurs in a sensor. This direct comparison is also completely free of the complication of FujiFilm’s different standard of ISO rating, something which often thwarts camera to camera comparisons.

There can be no more direct a comparison than this. No false color suppression or noise reduction is employed. Bring on the magic!

Example 1

Example 2

Example 3

Example 4

Huh. I wonder what happened to the magic. Can you see the difference? I can’t. They all look crusty as a Leprechaun’s corduroys on St. Patrick’s Day. (Leprechauns are magical right?) Do any of these examples look “film like?” Let’s see what happens when we apply a little noise reduction to one of these. Let’s use a Bayer image, with its supposedly un-film-like characteristics…

Ah, there we go. Well, I wouldn’t call this film-like either, but at least the color noise is gone. This is the same thing that happens in FujiFilm’s in-camera processing and in 3rd party RAW processors which don’t allow the user to completely disable noise reduction for X-Trans files.

PSNR stats with noise

Example       Bayer       X-Trans       Winner
1             16.61       16.63         X-Trans
2             16.34       16.30         Bayer
3             16.03       16.13         X-Trans
4             15.31       15.31         Tie

Surprisingly, X-Trans does have the win here — albeit a rather marginal one. You can see for yourself in the images how much of a visual difference these few decimal points of PSNR correspond to. Even if we don’t call this a tie, differences of this order would be completely swamped by 8-bit quantization, JPEG compression, optics and other real-world factors. Poof.

Versus an AA Filter

AA-filterless sensors (or sensors with negated AA-filters) have become popular in part because their output appears sharper straight off the sensor, without digital sharpening. As we’ve seen, this comes with the cost introducing false color and moiré artifacts.

A sensor with an AA filter requires digital sharpening, but with it can appear almost as sharp (more on this below) as an image from an AA-filterless sensor while displaying fewer artifacts. AA-filterless sensors require little or no digital sharpening, but are subject to false color and moiré effects (which as we’ve seen require digital noise reduction filtering to suppress).

An AA filter, however, does nothing to help with noise, which happens in the sensor, so false colors introduced by high ISOs (as in the noise examples above) are unaffected.

Where AA-filterless sensors may have an advantage is when the final image is to be converted to monochrome, and especially when the subject itself is monochrome (i.e. documents). And to a lesser extent when photographing subjects which contain no patterning and no high contrast fine detail (i.e. some types of nature/landscape images). Also, there exists computational diffraction reduction technology (which may be what FujiFilm is using in their so called Lens Modulation Optimizer) which actually relies on aliasing to function.

In order to simulate an AA filter, we apply a blur filter to the high resolution target image, downscale it to the testing resolution, perform the RAW conversion and demosaicking, and then apply an unsharp mask filter to the output. No noise reduction or desaturation is required.

Here’s the image from Example 4, with a simulated AA filter Bayer sensor:

The AA-filterless example looks like it has more detail, but does it really? And at what cost? Is it real detail or just aliasing and false color? That superfine “detail” (which is really just aliasing) isn’t going to be visible when you zoom out, but the false color may, especially with examples of moiré.

And let’s take a look at that moiré target with a simulated AA filter:

AA-Bayer VS X-Trans

Now let’s compare our simulated AA-filter Bayer output to X-Trans. This was the original case for X-Trans, touted as a superior alternative to AA Bayer…

Conclusions

The Markesteijn algorithm does a better job of interpolating near-diagonal lines than does the AHD alogrithm, but this isn’t directly attributable to anything about X-Trans or Bayer; other Bayer demosaickers perform better in this regard, and other X-Trans demosaickers perform worse. (Add to this that AHD hasn’t been tuned for unantialiased input, while Markesteijn is doing extensive 3-pass luminance interpolation.)

However, even with the algorithmic lead, X-Trans appears to offer no advantage over AA-filterless Bayer, and in fact produced a lower fidelity result than Bayer in all but one test case. Predictably, AA-filterless Bayer and X-Trans suffer from similar levels of false color, X-Trans being slightly worse/chunkier due to the courser pattern. X-Trans tends to produce line-like artifacts, appearing smeared in aggregate, whereas Bayer produces more speckle-like artifacts. The X-Trans pattern changes the character of moiré, but does not appreciably reduce it, and certainly doesn’t eliminate it.

The real moiré and false color reduction of FujiFilm’s cameras comes not from the choice of sensor CFA, but from noise reduction occurring in the image processing pipeline. As shown here and in my previous articles, this level of post-processing, which is applied globally and indiscriminately, has the side-effect of significantly reducing fine color resolution. X-Trans is more sensitive to subject color, performing its worst on subjects with predominant red or blue hues. X-Trans provided a marginally higher PSNR than Bayer in the presence of noise (the results are so close that things like a different choice of Bayer algorithm, JPEG compression, and certainly any application of NR would wipe out the differences). Any apparent larger noise advantage found in other comparisons must be due the confounding factors of underlying sensor technology (Sony makes the sensors, FYI), ISO rating, electronic/thermal noise, and noise reduction baked in to the X-Trans demosaicking algorithm in use.

Even though X-Trans lost the battle, the results were very close. An AA-filterless Bayer sensor and an X-Trans sensor of the same resolution are fairly evenly matched. A sensor with an AA filter, however, will beat them both hands down when it comes to eliminating moiré and false color — and without reducing color resolution in the process (but one must apply the appropriate amount of sharpening for the best results). This leaves us in a bizarre situation because FujiFilm insists on continuing to use X-Trans in their midrange/high end cameras (except for the medium format GFX 50S), while using Bayer (with AA-filter of course) in their low end X-A range. That means you can get better IQ by buying a low-end camera (X-A3) that costs a third what the high end model (X-T2) does — from the same manufacturer (although you won’t necessarily get better JPEGs if the level of chroma NR being applied in-camera is the same in the X-A line).

True AA-filterless sensors (where the AA filter isn’t simply negated) might have a slight physical sensitivity advantage due to receiving light which would have otherwise been absorbed or scattered by the AA filter. However, given the amount of noise reduction required to treat the false colors introduced by the omission of the AA filter, it seems unlikely that there is much, if any, net benefit.

Finally, and probably most practically, X-Trans requires significantly more processing time/power and, at the time of writing, all but one of the commercial RAW processing programs on the market produce lower quality output than the free-software Markesteijn algorithm used in preparing the examples for this article.

So there you have it: We disenchanted the marketing copy, saw through the legerdemain, traced the rainbow back to its very source, and found the truth about X-Trans. And that truth is: all the “magic” is nothing but smoke (sans mirrors). But, I have to admit, it’s a clever trick.

Jonathan Moore Liles is a photographer, writer, musician, and software architect living in Portland, Oregon.

website / instagram / bandcamp

When FujiFilm’s X-Trans III sensor was introduced in the X-Pro2, many users began noticing a strange new artifact in their backlit photographs. Upon further experimentation, it became apparent that the same artifact could also be found in images from cameras using the older X-Trans II sensor.

Many theories have been bandied about in internet photography forums, pointing the finger at specific lenses, certain body production batches, and, sadly but predictably, the users who dared to suggest there could be flaws in the output of a rather expensive camera, but very little information of a technical kind has been published on the problem.

In this article, which is the third episode in this X-Trans saga (the first being about color detail in the camera JPEGs and the second about luminance detail in RAW processing), I will share some of what I’ve uncovered about the nature of this particular artifact on my reluctant (some might say heroic) journey to become an expert in FujiFilm’s quirky and eccentric sensor technology. Come Sancho, we must slay this wizard who enchants the people’s sensors!

Joking aside, the first thing I’ve discovered about this issue is that it is very rarely encountered in practice by those who abstain from shooting facing into the sun or similar light sources. For those who indulge in flare-filled portraits or landscapes with the sun in the frame, however, it may be a more frequent occurrence.

Let me be perfectly clear: I’m not trying to play-up the severity of this problem by writing this article (it’s only happened to me a few times), only to offer some insight into its mechanism and cause, sharing what I have learned from many hours spent studying the issue in detail.

Bear in mind that this is a highly technical subject and this article will only scratch the surface of the issue. If you’re expecting a discussion on semiconductor fabrication techniques, electron beam coatings, etc. you’ll have to look elsewhere. The information presented here comes entirely from my own original eye-straining analysis of real-world images. I’m not claiming it to be 100% accurate. What I call “left” could be “right” etc. — there appear to be no authoritative reference materials published on the matter by FujiFilm or anyone else. (If someone out there reads Japanese and knows where to find the patents, by all means send them my way.)

The nature of the grid

This artifact is particularly interesting because it allows the layout of the X-Trans CFA to shine through, as it were, in the demosaicked image — something which should never happen. (If you look closely you can make out the “X” of X-Trans — uninterrupted diagonal lines of green pixels which criss-cross the sensor.) Not even the camera JPEG output, generated with FujiFilm’s supposedly expert proprietary image processing, is immune to this problem (And, yes, I’ve confirmed that Iridient isn’t immune to it either.)

Due to the complexity of X-Trans processing, the appearance of the effect will vary with the particulars of the demosaicking algorithm in use, but no algorithm will be completely immune from its effects. (It may be possible to include special measures to mitigate this artifact in a new algorithm, but this would further increase the complexity and computational load, and come at the cost of resolution and the introduction of new types of artifacts.)

Why is there a grid?

First and foremost, the reason that this effect is apparent at all is because the of the particular arrangement of the X-Trans CFA, with larger gaps between same-colored pixels. If a sensor utilizing a Bayer CFA were similarly affected, the presentation would probably be more like speckling than a grid, and certainly wouldn’t show any X’s, and could more effectively be removed by traditional noise reduction techniques.

What causes the grid to appear?

The X-Trans II sensor, found in the X-T1 (but introduced earlier in the X20, X100S, and X-E2), was the first to bring on-sensor phase-detect autofocus technology to FujiFilm’s X series of cameras. X-Trans III, found in the X-Pro2, X-T2, X-20, and X100F, extends this concept with a larger coverage area and more phase detect pixels (PDPs).

FujiFilm’s technology adds an additional layer to the sensor, a masking layer between CFA and the photodiodes. This mask is only apparent in the central region of the sensor (the extent being greater in X-Trans III than X-Trans II). It should be noted that there are many more masked PDPs on the sensor than there are “AF points.” 2.8% of pixels in the PDAF area of the sensor are masked. On the X-Trans III sensor, the PDAF area is 3000x3000 pixels (9MP), containing a total of 250,000 PDPs. AF points in this context are a software construct — the values of many PDPs are be used to determine the focus at a single AF point.

What this masking layer does is block half of the masked pixels from receiving light from the “left” side of the image, the other half from receiving light from the “right” side of the image. When the image at the AF point is in focus, the light from the two sides coincides (is in phase). Each PDP is only receiving up to half the amount of light of an unmasked pixel (1-stop less in photographic terms). This can be compensated for by doubling its brightness in software, with the penalty of also amplifying its noise. This system is also subject to interaction between medium to high frequency detail in the image and the mask (particularly apparent in feathers and fur), but that’s another problem, for another day. This particular implementation of on sensor phase detect is of the “horizontal” type, meaning it is only sensitive to vertical edges in the subject. (Most DSLR cameras have AF modules which include a mix of horizontal, vertical, and, more recently, cross-type sensors. Being limited to horizontal only sensing is a limitation of all currently deployed on-sensor PDAF technologies that I’ve surveyed, and isn’t exclusive to FujiFilm.)

But what does any of this have to do with the grid artifact?

There are at two main factors in play, both of which seem to involve this pixel masking layer. The overall effect is a combination of these factors, the precise appearance of which depends on the particular angle/orientation of the flare and the region of the frame the flare covers.

The phase detect pixel effect

To put it simply, it is possible for extraneous light to pass through the lens and strike the sensor in such a way that most of the “left” (or “right”) masked PDPs are not illuminated (although everything else is).

Don’t believe me? Forget about optical inversion for the purposes of this thought experiment (it’s a superfluous complication): Say that a cone of hard light (flare) is shining on the sensor from the “right” direction. This illuminates all of the unmasked pixels, and all of the “right” masked pixels, but none (or few) of the “left” masked pixels. It really is that simple.

When a demosaicking algorithm (even FujiFilm’s proprietary one) attempts to construct a full color image, these shadowed pixels misguide the interpolation, spreading the error out over a wider area, and allowing the pattern of the CFA to show through. Because of the alternating pattern of “left” and “right” PDPs horizontally across the image and the 12x12 repetition of the PDP mask, this effect creates an artifact with a period of 6 pixels horizontally and 12 pixels vertically across central region of the image.

OK, but why is the flare purple?

If you’ve been paying close attention (particularly to the diagram above), you may have already figured that out: The flare isn’t purple, it’s anti-green. Purple, more specifically magenta, is the color you get in RGB additive color mixing when you subtract green from white. That is to say, a mixture of just red and blue. The flare appears purple or magenta because of all the thousands of masked off pixels on the X-Trans II/III sensor, every single one of them is a pixel sensitive to green light, and located in exact the same place in the CFA pattern (upper right hand corner of that block of four green pixels). When a (white) veiling flare illuminates all of the pixels except for either the “left” or “right” PDPs, this leaves a deficit of green signal. (Note that in the real world, flares do tend to have a color tint of their own, but that doesn’t change the principle at work.)

The masking layer thickness effect

The PDP influenced part of the effect only appears in the central region of the sensor where the masked PDPs are, but the purple flare/grid artifact affects the entire sensor. This effect seems to be caused by the added thickness of the masking layer or perhaps some other property of the sensor’s optical stack.

What appears to be happening in this effect is that light is striking the sensor from the “up” direction and casting a “shadow” from one row of pixels to the row below. This is presumably happening in the gap created by the masking layer, between the CFA layer and the photodiode layer.

Pardon the annoying animated GIF below, but this was the easiest way to visualize what’s going on. This animation comprises three frames: The first frame is the (naturally) monochrome RAW sensor data, the next frame is the raw sensor data with each pixel colorized to match the X-Trans CFA pattern, the final frame is the demosaicked image data (where the grid and purple color can be seen.) This is from an area of the image which would have been uniformly dark (shadow) were it not for the flare.

As you watch this animation, pay particular attention to the top two pixels in each 2x2 block of green pixels. In the row below, you can see the intensity level that those pixels should have, notice how they’re darker, and that the green pixel below a red pixel is a different shade than the green pixel below a blue pixel? Can you also see that all the blue pixels immediately below a red pixel are darker than the blue pixels below a green pixel and vice-versa? The green pixels don’t appear to cast any kind of “shadow” in this way, only the red and blue pixels do. (Perhaps because the green filter is weaker or because of color shifts caused by the various coating involved or some other effect— the physical particulars don’t really matter at this level of analysis.) This pattern affects every 3x3 group of the X-Trans pattern, and repeats on a 3 pixel period horizontally and vertically across the image, creating the bulk of the “grid.”

OK, but why is this one purple?

It should be obvious from referring to the figure illustrating the X-Trans CFA that every third row of X-Trans has an equal number of red, blue and green pixels. That is to say, it is 33% green. The remainder of the rows are 66% green.

When a 33% row casts its “shadow” on the 66% green row below it, it is removing a significant amount of green signal from the image (the image of the flare, that is) simply because the 66% green rows have a larger contribution to the green channel. This isn’t even accounting for the fact that none of the green pixels appear to cast this “shadow.” This minus-green effect results in the flare appearing magenta overall. All told, this “shadowing” effect is responsible for the majority of the magenta tint.

Examples

Well, since you’ve read this far, I guess I’d better show you some examples. Unfortunately, Sancho and I were unable find any conveniently located windmills (trust me, there are at least a couple of dozen people on the planet who will find this joke mildly amusing), so this plastic flamingo lawn nativity scene will have to do.

You may think that the example image below isn’t a good one. Please try to bear in mind that the purpose of the example is to show the purple flare/grid artifact in a real-world context, not to present a composition for artistic criticism. You may be tempted to point out that the image is out of focus, and think that this somehow invalidates the example. It does not. Indeed, it may very well be out of focus, and if you’ve been following closely you will know why: flare (and to some extent any backlighting) causes the on-sensor phase-detect autofocus system of these cameras to go haywire. The camera doesn’t know what’s in focus. It’s hopeless. I suspect that in-focus examples of this problem are the exception rather than the rule.

The image below was shot with the FujiFilm X-T2 using the Fujinon 35mm F2 lens at f/4.0 ISO 200.

What can be done about it?

Unfortunately, not much. From a software perspective, you could insert some preprocessing before the demosaicking algorithm to identify the flare area, add some of the value of the red/blue pixels to the green pixels in the rows immediately below them (assuming the flare usually comes from the “up” direction), thus compensating for the masking shadow. In addition, you could have a demosaicking algorithm that ignores all of the PDPs, interpolating around them. That would probably get rid of the grid for the most part, and the purple aspect, but doing so would come at a cost to resolution, in particular the green/luminance resolution, an extra quantity of which was supposed to be the saving grace of the X-Trans CFA. This would all be absurdly complicated for a demosaicking algorithm and likely to introduce some new artifacts.

A hardware solution would be to ditch the current method of on-sensor PDAF in favor of something more sophisticated like Canon’s Dual Pixel AF technology (with which such imbalances as described herein are presumably impossible because there is no masking layer and no lost light). No camera or lens yet designed can perfectly reject flare; this problem is less about the flare occurring, which is inevitable under the described conditions, and more about the way the sensor responds to the flare.

It’s worth pointing out that all of these problems could have been anticipated by FujiFilm’s engineers before ever coming close to the manufacturing stage — they just didn’t think it was a big deal. Given that they went on to release three more camera models with the same sensor design after the problem was discovered by the public, I wouldn’t hold my breath waiting for them to issue a recall over it.

Conclusions

It is obvious from the single-pixel extent of the artifact in the raw sensor data that this is a sensor-level effect. The grid/purple flare is not due to internal reflections between the sensor and the lens (although this kind of reflection certainly can and does happen with mirrorless cameras), but to optical or electrical effects occurring within the sensor package itself. Any precautions to avoid or eliminate flare may reduce the the symptoms, but the disease remains. The underlying problem is exacerbated by presence of the X-Trans CFA, which imparts both the grid-like luminance effect, and the majority of the magenta colored chrominance effect.

As can be plainly seen, the overall effect isn’t particularly noticeable at typical (at the time of writing) web display resolutions. The purple tint is present at all display sizes, whereas the grid requires magnifications higher than about 25% to become apparent. However, the grid, consisting of high frequency detail, is subject to enhancement by sharpening and other post-processing steps, which may increase its visibility at lower resolutions. Whether or not you consider an image with this artifact to be completely ruined is entirely up to you. (Many people consider an image with any degree of flare to be ruined.) However, this is definitely a lower level of fidelity than I’m accustomed to seeing in similar situations. Furthermore, as already mentioned, due to the mechanisms involved, it is likely that the grid artifact and phase detect AF failure are, shall we say, comorbid and linked.

This artifact is characteristic of the FujiFilm X-Trans II/III sensor, allowing affected images to be easily be identified. I can’t recall another instance of such a complex and distinctive artifact. It is, however, easily avoided by abstaining from photographing backlit subjects.

Is this mere tilting at windmills? I don’t believe so. The problem is real, if infrequently encountered, and having an understanding of its nature can help us avoid it.

☂

Jonathan Moore Liles is a photographer, writer, musician, and software architect living in Portland, Oregon.

website / instagram / bandcamp

In my previous article, X-Trans: The Promise and the Problem, which focused on the difficulty of demosaicking FujiFilm’s X-Trans sensor data while preserving fine color detail and in particular the trouble FujiFilm’s own image processing pipeline has with it, I used the Free/Libre software Darktable to process the RAW examples. I showed that, specifically in terms of color detail, Darktable was able to do a better job than FujiFilm’s own processing. (But I also pointed out the compromise between color detail and false color/moiré inherent in X-Trans.) Several commenters suggested that I could get better results from a commercial software product called called Iridient Developer (which, it should be noted for those yet unaware, cannot be installed into your FujiFilm camera’s firmware in order to improve its JPEG output).

Iridient has become a popular alternative/adjunct to Adobe’s Lightroom for FujiFilm X-Series camera users who wish to process RAW files, due not so much to Iridient’s excellence as to Lightroom’s inadequacy in desmosaicking X-Trans images. (I don’t have access to Lightroom and therefore cannot provide my own example of its output, but a simple web search will provide you with more than you need to confirm this assertion.)

Admitting that if we want to preserve fine color detail we must abandon the camera JPEG output, the question remains: which raw processing software will provide the best results? While everyone seems to agree that Iridient is better than Lightroom, the question has never been answered as to whether or not it is better than Darktable.

Since this is clearly a subject still mired in confusion, I thought it would make an interesting topic for another article.

I’ve seen the results of several shootouts between Lightroom, FujiFilm JPEGs, and Iridient, and I know from my own experience that Darktable’s output is very similar to Iridient’s (as exemplified in the aforementioned), but I haven’t come across any direct comparisons — which isn’t very surprising considering the fact that Darktable is not a commercial product and nothing is to be gained financially from promoting its use.

In this article, we will explore the differences and similarities as they relate to image quality, in particular the quality of luminance detail.

Note that Darktable, dcraw, UFRaw, RawTherapee, and perhaps other Free Software RAW processors, all use Frank Markesteijn’s algorithm for demosaicking X-Trans images, so similar results can be achieved with any of them, but we’ll focus on Darktable here because it is, in my opinion, the most capable and mature, and the program I use the most personally.

Examples

Buckle up and get ready for some crops! Put on your reading glasses if you need them. Have a cup of coffee (double espresso for me, thanks). Even with these prerequisites, I’m sure some percentage of you will be squinting at the screen and wondering what the heck I’m talking about. That’s OK. Visual acuity varies as widely as does opinion on the significance of detailed image analysis.

Methodology

Iridient output was generated by Iridient X-Transformer (which produces a demosaicked RGB DNG file [and not a re-mosaicked Bayer image as is commonly believed]) with the following settings: Detailed, Sharpening Off, Luma NR Off, Chroma NR Off, Lens Correction Off. This image was then post-processed through Darktable, applying the same color profile, basecurve and sharpening (none) as the other images to permit pixel to pixel comparison.

Darktable’s output was generated by processing the RAF file directly with the Markesteijn demosaicking algorithm (3-pass mode), 2 iterations of false color smoothing (chroma median filter — this amount selected to match the appearance of false colors in the Iridient image), and no sharpening.

It is apparent that Iridient is applying a small amount of sharpening even with sharpening disabled. It took an unsharp mask of radius 1.5 amount 0.15 applied in GIMP to get the Darktable output to match visually (matched output pictured). This level of sharpening is insufficient for optimal viewing of these images at 1:1 (so be aware they’ll look a bit blurry here), but so many readers were confused by the use of sharpening in my last article I decided to try this one without it.

Example 1

ISO 200 image from the X-T2.

Having trouble spotting the difference? Let me help: here’s the difference image:

Still can’t see it, huh? Well, that’s because there’s not much difference to see. OK, stop fiddling with your monitor settings, let’s scale the whitepoint on that from 255 to 25 so we can actually see what’s going on down there:

What you’re seeing are subtle differences in the appearance of false colors and the effect of the different sharpening filters (unfortunately I could not find a way to completely disable sharpening and NR in Iridient). Iridient also seems to be using some a edge aware mean or other smoothing filter on the luminance channel even with luma NR disabled (which Darktable isn’t doing).

If you squint you can see a few single pixel errors Darktable’s output where some of the the X-Trans III senor’s masked phase detect pixels are located. Iridient must be compensating for these pixels (as it should), perhaps just incidentally due to the luminance NR it’s always applying.

For comparison, here’s the camera JPEG output:

Conclusions

The algorithms employed by Darktable and Iridient produce extremely similar output (so similar, in fact, that they may even be the same underlying demosaicking algorithm).

Iridient applies some additional post processing which cannot be disabled (or is perhaps using a joint-demsoaic/denoise algorithm), while Darktable leaves absolutely all noise reduction, sharpening and other filtering to the user’s discretion (which is quite easily verified by reading the source code). However, Darktable is not interpolating around the phase detect pixels as it should for this sensor technology (although these single-pixel errors are unlikely to be noticed by all but the most critical viewers. [i.e. apparently just me]).

For all practical purposes, the output of the two programs is identical, even under close 1:1 inspection.

What’s more, Darktable et al. are free to use, share, and modify — offered to you from the goodness of their respective developers’ hearts.

I know what I’m going to use.

☂

Jonathan Moore Liles is a photographer, writer, musician, and software architect living in Portland, Oregon.

website / instagram / bandcamp

FujiFilms’s X-Trans III sensor has been out since the X-Pro2 hit the scene in March 2016. It was joined recently by the X-T2, and presently the X-T20 and X100F are about to be unleashed upon the world in February 2017. I’ve spent some time with both the X-Pro2 and X-T2, as well as every generation of X-Trans sensor package going back to the X-Pro1. In the process of using these cameras, I’ve become intimate with X-Trans CFA and the problems it presents.

With this new wave of cameras on the verge of release, I’ve decided to share some of what I’ve discovered.

X-Trans is, for the purpose of this topic, FujiFilm’s alternative color filter array for CMOS sensors.

FujiFilm claims that:

“Moiré is tackled at its root cause by the revolutionary X-Trans CMOS sensor’s colour filter array. By enhancing aperiodicity (randomness) in the array arrangement, the colour filter minimizes generation of both moiré and false colours, eliminating the need for an optical low-pass filter in the lens[sic] and enabling [the] X-Trans CMOS sensor to capture full “unfiltered” lens performance.”

FujiFilm has also claimed that X-Trans provides a more “film like” image than the much more common Bayer CFA and can produce as good or better IQ than a full frame Bayer sensor (with AA filter). All of these statements are… shall we say redolent of a certain barnyard aroma? What X-Trans really does is trade a bit of chrominance resolution for a bit of luminance resolution (which you may love if you shoot Black & White) and make the demosaicking process (the interpolation of the RAW sensor output into an RGB image) far more complicated. There’s nothing random about the X-Trans CFA either, it’s an ordered pattern that repeats on a slightly larger scale than Bayer (6x6 vs 2x2). And recently most manufacturers have ditched the AA filters on their Bayer sensors as resolution has increased anyway, removing the original point of distinction. All of this applies whether you shoot RAW or JPEG, but Fuji’s JPEG engine has some particular shortcomings. Hey, that’s marketing for you. Now let’s try to ignore that smell on our boots and move on.

X-Trans JPEGs are “Waxy” at High ISOs

FujiFilm has been praised by many for their “color science.” That is, the reproduction of colors and tones in the straight out of camera JPEGs, through what FujiFilm calls Film Simulations. For this reason and for the generally high quality of the in-camera demosaicking vs. some infamously half-hearted attempts by Adobe et al., many photographers have announced that they prefer to eschew RAW processing and just deliver the SOOC JPEG images instead.

However, many who have attempted this have run into a significant obstacle, commonly known as the Waxy Skin-Tone problem. Much has been said about this issue elsewhere online, as the problem has existed at least since the introduction of the X-Pro1 in 2012.

I’m going to attempt to offer some evidence and insight into the issue as it applies to the current generation of cameras, and, in conclusion, ask Fuji what they plan to do about it.

Breaking it down

I mentioned earlier that the X-Trans CFA strikes a different compromise between luma and chroma detail (and noise) than Bayer. As you can see from the above figure, the X-Trans array has only 88.89% of the chroma resolution (red and blue photosites) of the Bayer array (which is traded for green photosites for increased luma resolution), and with larger gaps. This is certainly part of the problem, but it’s not the whole story. The demosaicking process is an interpolation process, whereby the colors missing from the sensor data are arrived at by an algorithm’s educated guess. With Bayer, and especially in the presence an optical low pass (AA) filter, the uncertainty of the red or blue value of a green pixel in the CFA has a specific limit. With X-Trans, this uncertainty is higher in part because the distance between same-colored pixels in certain directions is greater, up to 6 pixels (this is important for interpolating across gradients)! In other words, at small scales the color of individual pixels in in-focus areas of the final image is more guess and less fact. This kind of uncertainty (with both Bayer and X-Trans) results in a type of artifact known as “false color” at the output of the demosaicking algorithm with certain subject matter. False colors are the symptom of incorrect guesses based on the limited information contained in the raw image samples. There are various techniques for mitigating this artifact, and new methods are being researched all the time. False color suppression and chroma noise reduction can, in some implementations, be treated with the same processing step, as part of the demosaicking.

It’s anyone’s guess what algorithms FujiFilm’s cameras use to demosaic and denoise images, but FujiFilm’s implementation performs similarly enough to known and documented algorithms that it’s not necessary to know the all of the details to understand where the problem lies.

Shooting at higher ISO reduces the signal to noise ratio of the image and exacerbates the problem of false colors (you may have noticed this before as colored blotches in high ISO images). In order to mitigate this, FujiFilm cameras ramp up the chroma denoising along with ISO, as do cameras from other brands. But, as mentioned earlier, with the X-Trans CFA, false colors are more of a problem than they are with Bayer, and stronger filtering is required in order to smooth them away. The problem of waxiness arises because Fuji decided to use much stronger chroma NR than is strictly necessary to suppress false colors in the general case. The result is that colors bleed together (especially red/blue hues). The effective color resolution is lower than it should be, even taking that 88.89% figure into account. Teeth and eyes become the color of the surrounding skin. Rosy cheeks appear wan and corpse-like, and, generally speaking, people are rendered cartoonishly.

Contrary to popular belief, the “NR” setting in the camera’s menus does not significantly affect this chroma smoothing at all — It only impacts luminance noise reduction.

Some viewers are reported to actually prefer this effect, but technically it is an objective and measurable flaw — the rendered image is no longer representative of the scene (known as the “ground truth” in academia).

The issue becomes a significant obstacle because FujiFilm cameras give the user no control in the matter. There is no “High ISO NR” menu setting like cameras from other manufacturers have. There is no separate chroma NR setting. You either live with the reduced quality of the JPEGs or you don’t.

The alternative is to shoot RAW. Which is fine, but demosaicking X-Trans files is less efficient than demosaicking Bayer files —and, as anyone who has tried it knows, this translates to a much slower workflow — and you lose all of that “color science” too, because, sadly (shamefully, in my opinion), FujiFilm does not publish color profiles for their sensors nor embed the color matrices in the RAW files as some other manufacturers do. And if you want to use the camera’s WiFi feature to transfer the images to your smartphone and process/post them on the go— you can only do that with the JPEGs.

Examples

Because the problem is one of color resolution, you are unlikely to notice the Waxy Skin-Tones problem in headshots or the like without involving high ISOs. In those cases there are hundreds pixels whose common values overwhelm the noise and allow the color of the ground truth to show in the final image.

The problem appears in fine detail, and is particularly noticeable in human faces at a distance, but also in headshots in the capillaries in the eye or the color of fine hair (where it differs from skin tone).

I have come across several real-world instances of this problem and have decided to present these rather than some kind of laboratory setup to illustrate that it is indeed a problem encountered in practice and to make it perfectly clear what information — what objective quality — Fuji’s JPEG engine tosses out as if it were noise. Whether you care about this discarded information or not is up to you.

The following examples are heavy crops. The images are cropped to illustrate the differences for viewers with all sizes of displays. Keep in mind that FujiFilm sold us a new 24 megapixel sensor on the promise of being able to crop more. Once you’ve noticed the effect, I think you’ll be able to spot it in uncropped images too.

All examples were shot on Fuji’s new flagship camera, the X-T2 with the Fujinon 35mm F2 WR lens.

A note about FujiFilm’s JPEG output: “FINE” quality in camera translates, in more specific terms, to 99% quality level JPEG with sampling factors 2x1,1x1,1x1. (This means that there is less color resolution in the horizontal plane than the vertical, but don’t get hung up on that because much of the color information in these images is interpolated by the demosaicking process anyway and it can be shown that this level of chroma subsampling of the JPEG image cannot account for the waxy skin tone effect [this will be left as an exercise for the reader.])

Example 1

The image was shot at ISO 1600.

Fuji JPEG

Note that the skin tone is very uniform and its color has bled into the sclera of the eyes, giving the face a waxen, wooden aspect.

Darktable RAW

The RAW file was processed using Darktable’s Markesteijn demosaicking algorithm (3-pass mode) with a single iteration 9x9 chroma median filter followed by application of a bilateral filter on the chroma channel and light sharpening. The color profile is my own, generated from shots of a Wolf Faust IT8 chart and should accurately represent the colors in front of the camera. No lens distortion correction was applied although the lens used (Fujinon 35mm F2) is heavily corrected in camera (which makes a more direct comparison difficult).

Note that the sclera are white. Also note the greater contrast and tonal variation present in the face, with a red nose and rosy cheeks. This is closer to the ground truth than the camera JPEGs, but not perfect (more on that later).

The Effect of the In-Camera NR Setting

Here is the same image as above, processed in camera at the extremes of the NR setting.

As you can see, the difference is extremely subtle and does not appear to affect the color/waxy skin tone situation at all. Just to drive this point a little further home, since there is much misinformation about this on the web, here is a difference image of the NR -4 and NR +4 images, blurred (3x3) to reduce the influence of JPEG artifacts, and stretched into a visible lightness (moving the whitepoint from 255 to 10):

I know, it’s not much to look at. You can see a little bit of difference in luminance information around edges and some subtle color variations (likely artifacts).

Now here is the same process applied to the processed RAW above, and the same image with a bilateral filter applied with a strength tuned to match the appearance of the camera JPEG. In other words, this is more or less what we could expect to see in the above difference image if the camera’s NR setting significantly affected chroma denoising:

Unfortunately, the inability to disable software lens correction in the in-camera RAW developer (not to mention the secret color profiles and LUTs of the Film Simulations) prevents a more direct comparison between the camera JPEG and Darktable’s output, but this should give us an approximate picture of the level of color detail being discarded by the FujiFilm JPEG engine. Note the differences in the eyes and tip of the nose, this is where the effect was most noticeable in this example.

Example 2

The image was shot at ISO 1600.

Fuji JPEG

Here we see once again that the sclera are skin-toned and the teeth have taken on the color of the surrounding skin. This is the epitome of ‘waxen’ appearance.

Darktable RAW

While in the processed RAW file, we see the true tooth color, the whites of the eyes, and even the red of the water line area.

Example 3

Remember how earlier in the article I implied that the Waxy Skin-Tone problem only cropped up at high ISO? Well, that’s only if you don’t look too closely. With X-Trans II/III, chroma NR is too high at all ISOs, from 200 up. It’s just more noticeable with the higher ISOs. But I don’t expect you to take my word for it. Here’s another example.

The image was shot at ISO 200.

Fuji JPEG

Note that the flesh visible on the head is an even hue.

Darktable RAW

Note the variation in the hue and saturation of the flesh on the head, which was bright red in reality.

Conclusion: A Devil’s Bargain

The X-Trans sensors and the X series of cameras have been plagued by image quality problems from their inception and it’s clear from these examples that the problem persists in the current generation (X-Trans III), whose members include the X-Pro2, X-T2, X-T20 and X100F. With every new camera generation, people claim that the problem is solved once and for all, but the reality is that it has never been solved, just tweaked here and there. The underlying problem remains.

On the bright side, it has been shown here that much more color information can be extracted from the raw sensor data than FujiFilm’s in-camera processing is currently capable of. Enough, in my opinion, to reasonably live up to the resolution claim of 24 effective megapixels.

So, why, you ask, does FujiFilm use such strong chroma NR if it causes this ugly rendering? If you’ve been paying attention you might have already guessed. The answer is simple: To suppress moiré! Ah ha, you say, but, according to FujiFilm’s marketing department (there’s that smell again), X-Trans is immune to moiré! Well, I’m here to tell you that it’s no more immune to moiré than Bayer and in fact probably a little less. It all comes down to the signal processing, and in this case, that processing is not without its side-effects. I can tell you with certainty that the level of chroma NR used when processing the example RAW images, which was mild enough to preserve the skin coloration, would barely make a dent in the waves of green and magenta false colors that come along with moiré. Don’t believe me? Go back and look at the man’s tie in Example 2 again. In reality the tie was gray — not covered in green and magenta glitter, just plain old shades of gray (and that’s not even a severe example). But recording that is a challenge for a sensor without an AA filter, and is even more challenging when that sensor uses the X-Trans CFA with its broader spacing of same-colored pixels.

X-Trans I had more susceptibility to moiré (still skeptical? go back and look closely at those camera settings images above, they were shot with an X-Pro1 and you can see that they exhibit some moiré and maze artifacts), X-Trans II swung in the opposite direction and introduced people to the full side-show horror of the Waxy Skin-Tones effect, and X-Trans III just tones it down little, striking a balance still in favor of protecting against moiré.

I say X-Trans N throughout this article, but what’s really relevant is not the sensor itself, but the image processing pipeline in each camera generation where the demosaicking and noise reduction happens.

The reason that FujiFilm hasn’t fixed the Waxy Skin-Tone problem after all these years and three camera generations is simply that they can’t — not without a significant, breakthrough advancement of their algorithms. They made a proverbial deal with the devil with their immune-to-moiré claims for the X-Trans CFA. If they get rid of the moiré, people will complain about waxy skin tones, and if they get rid of the waxy skin tones, people will complain about moiré (but, hey, they never made marketing claims about not making people look like wax figures or wooden dolls, so it’s no surprise which problem they’ve prioritized). Certainly, more sophisticated moiré suppression algorithms could do better at sparing skin and faces where the color smoothing is most objectionable (but could they ever recognize chicken skin, I wonder?) Will Fuji ever invest any time or money in that kind of optimization? Especially with reviewers constantly praising their JPEG engine (from a comfortable distance and without wearing their spectacles)?

In the future, I sincerely hope that FujiFilm stops producing sensors with the X-Trans CFA. I am convinced that the X-Trans CFA causes more problems (many) than it solves (none) and FujiFilm could have well known this from computer simulations before ever manifesting it physically. If the GFX 50s is any indication, then the next generation of X-series cameras may indeed utilize a Bayer CFA.

In the meantime, a firmware update to give users the option to customize the chroma NR/moiré removal strength ramp ourselves in camera (like some other manufacturers like do with their High ISO NR customization) could alleviate the problem for users who consider the lifelike rendition of human faces more important than complete freedom from moiré. I challenge FujiFilm to offer a solution. Any solution. Because until they do so, the JPEG images from your new FujiFilm camera are going to be compromised in color resolution. For this reason I would strongly recommend recording RAW files in addition to JPEG if you care about this sort of thing.

All of that being said, allow me to put things in perspective by sharing the uncropped image from Example 1, both the camera JPEG and the more mildly denoised RAW (via Darktable). Whether you notice this effect depends on how much you crop/how big you print and how high you push the ISO. It has annoyed me enough that I’ve become adept at spotting it and wasted a couple of perfectly good evenings writing this article. When I look at an image and the teeth are skin-toned or the veins in the eyes are gray, I can confidently say to myself, “That’s a FujiFilm!” And perhaps, for better or worse, now you will be able to do the same.

Addendum 1

Some readers were curious why the Camera JPEG examples used Color +4 and Sharpness +4 settings. This was done in the interests of fairness and clarity. Fairness because the Provia Film Simulation (the Fuji standard value) is quite desaturated compared to the calibrated RAW data, and this desaturation makes the lifeless appearance even more dramatic. Clarity because it’s a crop, crops are less sharp, and it’s uncomfortable to stare at unsharp images. Here’s the image from Example 1 with full default settings (all zeros):

Addendum 2

Some readers were unsure what I was referring to when I mentioned the X-Pro1 and maze artifacts. Here’s a crop of one of the images referred to. Yes, I know, it’s a crop, but now you see what I’m talking about. Deal with it.

Addendum 3

A reader asked to see the uncropped images from Example 2 and 3. Not exactly Pulitzer Prize material here, but, hey, you asked for it.

Addendum 4

A reader has complained about my opening paragraph, saying that FujiFilm has never claimed that the X-Trans CFA is random, only that it is “more random” or has “increased randomness.” My position is that this wording alone constitutes a very misleading statement, but they have indeed used stronger wording.

Direct from the FujiFilm website regarding the X-Pro2’s X-Trans III sensor:

“The unique random color filter array reduces moiré and false colors without an optical low-pass filter. These color filters also have the effect of increasing the resolution so, when shooting with a high-resolution Fujinon lens, the camera delivers images with a perceived resolution far greater than the actual number of pixels used.”

I’m not going to present any more examples of this language. Interested readers can peruse the web and satisfy themselves that this is the way FujiFilm presents their technology. There are many of other instances both written and spoken.

☂

Jonathan Moore Liles is a photographer, writer, musician, and software architect living in Portland, Oregon.

website / instagram / bandcamp