As I noted in my earlier post, the X-trans sensor in the Fuji X-series cameras seems to have amazing noise performance even at high ISO.. to a degree that seems unreasonable. Apparently, others have noticed the same thing, and a plausible explanation is that the camera is applying noise reduction when generating the RAW image file. It’s hard to make general conclusions here, but this worries me. 

How do we know noise reduction algorithm isn’t making an unwanted tradeoff in the quality of the image? Especially since a PC has way more processing power than a camera does, such a strategy by Fuji forces the user into a single processing approach, which is against the spirit of the raw workflow. To wit, the link referenced above says noise performance looks pretty good up to ISO 1600, but above that, it becomes splotchy and the reviewer prefers the more “natural” noise look of the Nikon FF camera. 

In theory, such an approach could be justified if the camera has some information that a later processing step would not have, and access to this information resulted in a theoretically superior result than anything a post processor could generate. But this seems unlikely, as there’s no reason that the camera couldn’t record this extra information in the RAW file and a post processor would only need to simulate the same process. There is a remote possibility that such “information” is of such volume that it would be impractical to record for every shot. But this also seems unlikely, for if this were the case, the processor in the camera would have also be powerful enough to munge through this extra data. 

The other implications are clearer: there’s nothing special about the electronic setup of the X-Trans sensor. It’s probably seeing just as much per-pixel noise as other APS-C sensors with the same number of pixels. It may have some algorithmic advantage due to the lack of a AA filter, or its funny non-Bayer filter arrangement, but any such advantage is being cooked into the output of the camera (JPG obviously, but RAW too).

I suppose there are legitimate reasons that a company might do this:

1. They want to keep their noise processing algorithms a secret, and what better way to do that than keep it in-camera.
2. They don’t want to have users rely on the post-processing software to implementing a very finely-tuned processing scheme, which would likely be suboptimal in some way, and thus make the product look bad overall. For a product that seems to pay such attention to it’s JPG output quality, this makes sense.
3. They want to game the reviews, where they know reviewers are going to be pixel peeping at results from both JPG and RAW output and comparing to other cameras. This in turn may sway potential buyers, who don’t know how to interpret the data.

Whatever the actual reason, this approach makes for some confusion when evaluating the X-series products against the competition. I suspect the behavior is surprising to most users of raw files, as the expectation is that the RAW format contains a most direct representation of what was recorded on the imaging sensor, with no processing applied. In the larger context of all the issues Fuji has been having with raw processor support, another small issue like this is unsurprising. It seems like Fuji clearly underestimated how important the RAW workflow was for their intended audience, and is now only making amends (e.g. rumors of working with Adobe et al for better raw support).

On a separate front, this set of articles linked from fujirumors.com was of particular interest. Chromasoft, a developer of some iOS-based raw processing software, delves deep into the x-trans sensor. Of particular note was the bit:

So my conclusion is, sorry to say, that the Fuji X-Pro1 X-Trans sensor doesn’t deliver the Fuji promise of outperforming similarly sized sensors. In fact, it underperforms similar DX sensored cameras – with the official SILKPIX raw developer, the underperformance is too slight to be noticeable under normal circumstances, but is still there if you look closely.

This is not to say that the Fuji X-Pro1 is a bad camera – far from it – the camera has great lenses, a really attractive viewfinder and design, and in many ways the sensor is great, with low noise and clean data, in the class of the recent Sony and Nikon sensors. If you use it with the official raw converter, it’s within a whisker of the competition. But in my opinion, it would have been a better camera with a conventional two-by-two sensor layout.

A pretty bold statement to say the least. On some level, I trust this guy’s opinion since he likely understands the intricacies of the demosaicing process way better than your average camera reviewer/blogger. On the other hand, he is just one guy, and on the other side, there is a team of Fuji engineers doing similar analysis and making explicit tradeoffs. It’s possible that the the X-Trans + AA-less design, while not resulting in improved resolution compared to other APS-C sensors, does confer other advantages, such as enabling a novel noise reduction technique, and/or as the marketing material says: avoiding moiré w/o the use of extra optical components.

Only time will tell, but I do hope that eventually Fuji come clean about what’s going on. If their RAF files have some noise processing applied, they should at least admit to the fact. It doesn’t necessarily make it a worse camera. iIn fact, given that Fuji has the greatest interest and making sure both NR and demosaicing work well for their special sensor, I’d be fine if the camera just generated TIFF files, where Fuji does all the work, but we still have the bit-depth to play with the images in post. It would be most interesting if the cameras were also modified such that RAF files really contained all raw data including any other parameters that are used as inputs to the hypothetical NR engine. This, of course, does not guarantee better images for all users.. without knowing the economics of the market, it’s hard to say for sure that raw processor developers would want to invest as much time as Fuji has to improve the output. 

For more information on the X-E1, check out my link collection.