Part 1 - Introduction
Part 2 - High ISO Noise
Part 3 - Dynamic Range

Many readers of this site are aware that, beginning with the F10 and culminating with the F30, Fuji has offered arguably the best high ISO performance yet seen in a compact camera. Professional reviewers and photographers alike attributed this ability to a combination of Fuji's proprietary sensor technology, the decision to stick with only 6MP, and expert application of relatively heavy noise reduction. Whatever the contributing reasons, the results were compelling. Here is an example of a resized ISO 1600 shot I took of my son in relatively low light (f/5, 1/38s):



With the F50fd and now the F100fd, a number of reviewers, photography blogs, and Fuji fans have complained about Fuji's decision to follow suit in the megapixel hike. Surely they cannot deliver F30-class high ISO performance with the F100fd, which crams 12 million sensels into a small (1/1.6") sensor, right? Yet Fuji claims that with the F100fd, "ultrahigh sensitivity combined with the next-generation of noise reduction captures the moment in beautiful photos like no other compact digital camera." The truth, known to many readers of this site (though not all or I wouldn't be writing this), is that high megapixel counts don't really impact noise performance the way they are commonly believed to do so.

Everyone knows that noise which is apparent when an image is viewed at high magnification is far less apparent when the image is viewed small. Put another way, downsampling an image improves the apparent level of detail relative to noise. Judicious and skilled application of software noise reduction and sharpening during the downsampling process can further enhance this perception of detail/noise. The opposite is also true. When we upsize images, the apparent detail relative to noise decreases. Comparing a 12MP ISO 1600 image from the F100fd to a 6MP ISO 1600 image from the F30, there is no doubt that the one with smaller pixels (F100fd) will have far more noise when both images are viewed at 100% (pixel level); however, the 100% view is misleading. One should take into account that in producing output of equal size whether onscreen or in print, the high MP image will require more downsampling or less upscaling, either of which will improve the apparent detail/noise in the F100fd image relative to the F30 image. A number of individuals, clued in to this concept, have demonstrated similar high ISO results with high MP compacts relative to the "F30/F31 standard." For example, look here for Prognathous' demonstration of a resized, noise-reduced crop from the 10MP Ricoh GX100 compared to the in-camera JPEG output of the F31.

In theory, for two sensors of equal size and technology (same high ISO read noise and light capturing ability), the pixel (more accurately sensel) pitch should not impact high ISO detail relative to noise performance whatsoever at a given output size. For technical discussion, refer to the many posts by John Sheehy as well as those of Emil Martinec on these matters. While the theory makes intuitive sense to me, I have to admit that in practice it doesn't quite work out. Consider the following small pixel pitch vs large pixel pitch comparisons: 1Ds Mk II vs 5D, G7 vs F30, 1Ds Mk III vs D3. In each case the generation of sensor technology is assumed to be reasonably close, and the sensors are of similar size; hence light gathering ability should be nearly matched. As expected, the small-pixelled cameras show far more "per pixel" noise at high ISOs than do the large-pixelled cameras. Resizing both images to the same size output, these differences are greatly mitigated. Yet from the examples I have seen, the large-pixelled cameras continue to hold a slight high ISO edge even after resizing. Some maintain that such differences come down to software processing (NR, sharpening, etc). I'll leave it to folks like John and Emil to clarify those particulars. The main point here is that the pixel-level view is highly misleading and that the commonly held belief that more megapixels means poor high ISO performance is, at least for the most part, unfounded.

Fuji has undoubtedly been working to improve their sensor technology, but even so it is highly unlikely that the 12MP files from the F100fd will be able to match the 6MP images from the old F30 when compared at 100%. Knowing well that the pixel-level view makes quite an impression and that most people want to have their downsampling and noise reduction done for them, they have enabled the F100fd to offer the ability to process in-camera JPEGs as 6MP files (similar to the F30). Unfortunately I do not have the F100fd in hand for testing. However, a Chinese blog at http://f100fd.seesaa.net/ (English translation by Google here) has quite a few 6MP high ISO F100fd images. Click here to see some of their high ISO F100fd samples. To my eye, they look similar to the output of my old F30 and a bit better than I can get from downsampling high ISO images from other high MP compacts. I don't know if the Fuji advantage comes from better sensor technology or more sophisticated on-board noise reduction software, but the results are impressive. Then again, if you're the type who likes to selectively and sparingly use NR, no luck. The F100fd does not offer RAW and won't let you disable NR on in-camera JPEGs.

Addendum: Prognathous has done some excellent side-by-side testing of the high ISO performance of the Fuji F31 and F100fd. You can find his comparison here. It seems that the F31 is still class leading in this area, though the F100fd has a respectable result.

In the final bit of this three-part look ahead at the F100fd, I'll discuss Fuji's claims for wide dynamic range. That disclaimer once again: I have no inside information from Fuji, nor have I had the priviledge to yet test the F100fd. What follows is simply a discussion of the known technology with a measured dose of good-natured speculation.

The new Adobe Lightroom update is said to have better noise reduction and sharpening tools than the previous version. From what I can tell, substantial luminance noise reduction (NR) is being applied to RAW conversions, even when luminance NR is set to zero. The result is a significant loss of fine detail and a "watercolor" appearance, even at low ISOs. If you want your Ricoh GX100 RAW conversions to look like in-camera Panaleica JPEGs, then Lightoom 1.1 may be the one for you. I'm really disappointed in this release. Hopefully I'm missing something here and someone will let me know how to disable this nasty noise reduction.

Addendum: It has been suggested to me that one can disable this noise reduction by switching 'off' the Detail section. I'll try this later and report back.

Addendum #2: That didn't work. I have been trying to adjust settings to overcome this problem, but I have been unable to do so. The best I can do seems to be turning down 'Luminance' noise reduction to '0' and holding back on the 'Detail' slider.

In following two screenshots, Lightroom 1.1 is shown on the right, with sharpening settings at default other than 'Detail' being increased to '100.' Luminance NR is set to '0'. On the left is a conversion using the older Adobe RAW engine with identical sharpening and NR settings. Click on the screenshot to view at full size. Some browsers may require a second click to get to 100% size.

Here is the Digilux 2 ISO 200 screenshot. Note how the character of the noise has been altered (ruined).



Here is the GX100 screenshot. The changes here are more subtle, but there is a definite loss of detail in the Lightroom 1.1 conversion. What concerns me though is not so much detail lost by LR 1.1, but the unnatural way in which the fine detail is rendered. The "unnatural" quality is no doubt worse by cranking up the 'Detail' setting, but the goal in this experiment was to approximate a similar level of detail between the two images.



Any oher thoughts or tips about how to overcome this issue? I can live with mandatory NR in a RAW converter. C1 has this as well. However from my standpoint, Lightroom 1.1 is too heavy-handed in its approach to NR at the minimum setting thereof.

Addendum #3: I've been wondering how this issue seemed to be getting so little attention. Thanks to rstockm for point out this thread in the Adobe forums, which contains posts from a number of photographers regarding this issue.

Below are two full-size GX100 files (about 5MB each) demonstrating the issue. They were both preocessed from RAW using the same default sharpening settings, except the 'Detail' setting was increased to '100' in LR 1.1 to try to match the detail of the other file. If you have a fast internet connection and would like to know more about the issue, try downloading them and comparing them in various regions.

Click here to download the image processed using LR 1.1.
Click here to download the same image processed using the previous ACR engine.

I recommend trying to process the two images linked above to see how they respond to further noise reduction, sharpening, levels/curves, etc. The "problems" (as I view them) become more obvious with any further manipulation.

Featured Comment by anonymous:

I'm seeing exactly what you're seeing, Amin. I saw it the second the first images opened in the new version of Lightroom. I hadn't noticed the problem in ACR 4.1., because I was processing everying in Lightroom 1.0. Yesterday, I ran files through ACR 4.1. and was shocked by the results.

I thought, at first, that only my small sensor work was reflecting the problem, but if I go through the imported Lightroom database there isn't a file that hasn't suffered as a result of the new algorythm - from 6 mp captures, through 10 mp, 12 mp and 16 mp.

Fortunately, I kept version 4 of ACR and version 1 of Lightroom, so with these back online I am able to work on. What shocks me is that the Adobe big guns, Thomas Knoll et al, sound as though they are satisfied with the new look. To me, it's the end of any pretence of photographic veracity for the digital image.

As for the new controls, if someone tells me that "Detail" is to put back in what was taken out, again, I'll lose it. And don't get me started on the "Clarity" slider. Have you seen some of the disasters being produced by cranking that Muvver to 100!? Good grief...

If the results of digital photography had always looked as they do after Lightroom 1.1. processing, I would have walked away from the photography business a decade ago. I am very worried that this 'new look' could become an accepted standard in our work. It's just damn wrong.