The first set of crops are at the default settings in the various raw converters. This typically includes some noise reduction and some sharpening. Exposure, contrast and color are set based on the software manufacturer's priorities. The original image is slightly underexposed and the crops are correspondingly a little dark. This is corrected in the post processed images. Note that there is no default image crop for PerfectRaw since the program makes no effort to interpret the raw data and the null settings produce a very dark and flat image. Indeed, Perfect Raw provides only a few adjustments, such as exposure and noise reduction. Its claim to fame is the ability to choose from eight different demosaicing algorithms, leaving most post processing to other programs. More on this later.

To my eye, C1P produces the best details without significant artifacts. Canon's own DPP comes in a close second and, not surprisingly, does a nice job with tonal values. In both of these, there are a few negligible sharpening artifacts visible in the tree branches. Surprisingly, the DxO crop lags behind in edge definition and shows a great deal of chroma noise. It does, however, do a nice film like tonal rendering and corrects for lens distortion, hence the slightly different crop. RT also lags behind in the definition of fine detail and shows chroma noise. ACR does a credible job, though it clearly blurs some of the fine, low contrast detail. At the default settings, we learn more about noise reduction and sharpening than we do about demosaicing algorithms.

Staying with the theme, the next set of crops are manually adjusted within each raw converter to produce the sharpest, most detailed output and still keep artifacts minimal.

Before discussing what I see in this set of crops, it will be helpful to explain what changes were manually made in each program. First, all luminance noise reduction was turned off. The ISO100 files from the 5DMKII are clean enough that with luminance noise reduction set to zero and sharpening set at a high level, no luminance noise is visible, even in the largest of prints. Second, chroma noise reduction was increased until chroma noise became insignificant. Third, chromatic aberration was manually adjusted and minimized. Finally, sharpening was set at a high level for fine detail, typically with the radius set to a very low value and the amount set near the highest value. This type of sharpening operates on the smallest scale and favors the finest details. The very detailed 5DMKII files hold up well to this approach, though its not appropriate for all cameras and images. The one exception to this approach was with the RT software, wherein there is an alternate form of sharpening known as RL Deconvolution. Without going into much detail about how it works, suffice it to say that it is an iterative process that is used in astronomy and, though slow, does a very nice on landscapes as well.

Having optimized the settings in each raw converter, we see that the differences in detail and edge definition are now smaller, though still visible. The least improved images are those from C1P and DPP, which suggests that the default settings are well optimized for this type of raw file. The most improved is the RT image. In fact, it may be the best of the lot. Looking at the second fence post from left, only the RT image clearly shows all of the metal ribs. Moreover, the tree branches are finely rendered and the grass shows more detail than all but the C1P image. The only downside I see is poor anti-aliasing in the darker, diagonal tree branches. These jaggies are less noticeable in the C1P crop, though there are other sharpening artifacts visible. In all, both the C1P and RT images are very good.

As an aside, Raw Therapee, like PerfectRaw, has the option to choose different demosaicing algorithms. Of the three available, an algorithm named HPHD produces the tightest, most even grain pattern, retaining the most fine detail in the final output. HPHD is the demosaicing algorithm used in this comparison. Combined with the unique RL Deconvolution sharpening method, the RT converter produces very detailed output indeed. This is remarkable considering this relatively well featured software is free. If you are looking for an ultimately cost effective raw converter or just like to tweak your images, give Raw Therapee a try.

Below are the crops, now including the PerfectRaw sample, with all noise reduction and sharpening turned off. For this comparison, the PerfectRaw AFD demosaicing algorithm was chosen because it retains the finest level of detail. This is the set of crops that most clearly shows the differences in demosaicing algorithms. It may also show that some noise reduction or sharpening is applied during the demosaicing process.

The PR crop clearly shows a different approach to raw development, largely limiting raw processing to demosaicing, without curves or color corrections applied. The demosaicing algorithm is also noticeably different than those used in the other raw processors, rendering by far the most fine detail. As you will see, there may be a cost to this method. The DxO crop again shows the most chroma noise and, to my surprise, the softest rendering of detail. To be certain there was no user error involved, I ran the test with DxO a second time and received the same results. Second only to the PR crop, the C1P crop again seems to retain very good fine detail. ACR, DPP and RT are similar enough that differences would not be noticeable in most prints.

Finally, the full set of crops below were post processed in Photoshop, with each image. receiving chroma noise reduction and identical amounts of sharpening. No luminance noise reduction was applied. As with the sharpening done directly within the raw processing applications, I used a very small radius and high amount. Specifically, the radius was 0.2 and the amount was 500%. While this may or may not be the ideal amount of sharpening for this image, depending on the desired output, it does give an apples to apples comparison of what can be expected of the various raw processors. Additionally, the PR image required the application of curves and color corrections to match the other images.

The first thing to notice is that the differences between most of these crops is now quite small, and likely insignificant in all but the largest of prints. Second, the relative pecking order has remained more or less consistent throughout the stages of this analysis. The exception is the RT RL Deconvolution sharpened image. Finally, the post processed versions of the C1P and PR images retain the most fine detail. The choice of demosaicing algorithms used in the PR image, however, produces poorly rendered diagonal lines. Notice the jaggies in some of the tree branches. C1P manages almost as much fine detail, yet maintains smooth, natural looking diagonal lines. In fact, throughout my experience using these raw converters, C1P consistently produced the most pleasing (for me) combination of detail versus artifacts. The question remains, however, whether the gains from any of these converters is significant enough to warrant a change of workflow from what I already know.

One thing is certain, well processed raw files produce images of much greater fidelity than are available from in camera JPEGs. In fact, the in camera JPEGs from the Canon 5DMKII show an unfortunate amount of luminance noise reduction at low ISOs, all but obliterating much of the fine detail. Should Canon ever add a feature in future firmware upgrades, the ability to turn off all low ISO noise reduction would top my list. Take a look at the crops below.

To RAW or to JPEG is no longer a question. Enough said.

In experimenting with some of these raw converters, I used mostly trial versions of the programs. It will take more experience with the various workflows and features to decide whether or not to make C1P or any other converter my go to software. In all, ACR does a good job of converting raw files and has a nice feature set. If you are using Lightroom, which uses the ACR engine, for organizing, viewing and creating output, it would be difficult to argue with the seamless workflow of using one program for everything, including integration with Photoshop. Capture One Pro produces top notch raw conversions, has an extensive array of tools, as well as organizing capabilities, and has really piqued my interest. Raw Therapee, used within certain parameters, produces remarkable output, has a surprisingly large number of features and will stay on my computer for certain tasks because it costs nothing to do so. As far as I can see, DxO Optics Pro gives you a couple advantages over the other programs, automation and the inclusion of specific lens corrections. The basic conversions, however, were a bit disappointing, though it may shine at converting higher ISO images. Canon's own Digital Photo Professional does a respectable job of converting raw files, but is lacking in controls compared to some of the other programs. Never-the-less, if you have a Canon high end camera, it came with the package and costs nothing extra to use. Finally, PerfectRaw is clearly a work in progress; indeed, in its current form it has not yet reached version 1.0. It does provide an opportunity to try several different demosaicing algorithms and can output the absolute most detail of any converter here tested, albeit with some artifacts.

In the end, the differences in image quality are subtle enough to be insignificant in all but the largest prints. This is particularly the case with the post processed set of images. The in camera JPEGs from my 5DMKII, on the other hand, will likely stay in the realm of web work and very small prints. I'm quite disappointed in Canon's approach to low ISO JPEG processing and quite pleased with how much detail is available in the raw files. Preliminary results from the new Nikon D3x show that in camera JPEGs can do a lot more.

I imagine that for professional photographers who work with very large numbers of raw files at once, workflow and automation will be paramount and absolute image fidelity secondary. For pros and amateurs who work with fewer images, workflow and automation might be more on par with, or even secondary to, image quality. Either way, photographers have individual needs based on their chosen subjects, equipment and final output. This article should therefore serve as an aide, not an end point, to finding the right raw conversion software.

This article has generated a fair bit of interest and feedback. The single most frequent question is "have you tried raw converter x,y, or z?" Since the list of available programs is long and time is short, I've decided to make the original raw file available for download. You may process and modify the image as you see fit for the purpose of evaluating photo software. If you post samples on another site, please be so kind as to give credit and provide a link back to here.

If you would like to have your processed samples added to this comparison, email them to me along with the software name(s), processing parameters and your name so I can give credit. I would be particularly interested in seeing output from Raw Photo Processor (RPP), a Mac only freeware program. To keep consistent with the methodology, I would need three crops 1) Use the program default settings. 2) Make manual adjustments for the most detailed output, including zero luminance noise reduction, moderate chroma noise reduction, correct for chromatic aberration, and very fine sharpening (typically a very small radius and large amount). 3) Same as number two with all noise reduction and sharpening turned off. Send only the crops (320x320 pixels) that match the samples already posted.

Image quality is a rapidly moving target in these days of the digital revolution. The constant improvements in hardware and software are welcome, though I feel endlessly compelled to rework my portfolio with every new generation of improvements. The latest, greatest update is the recently released beta version of Lightroom 3. Among other improvements, Lightroom 3 boasts a new raw processing engine. This is welcome news, as raw processing was by many considered a weak point of prior versions. Indeed, my own exploration into raw processors, resulting in this article, showed Lightroom 2 to run in the middle of the pack. My brief foray into Lightroom 3 beta shows that Adobe has made some remarkable improvements, possibly ending the need to have more than one raw processor on my computer.

In short, the new processing engine produces noticeably finer grain and detail from the Canon 5DMKII raw files. In fact, it equals the previous champ, Perfect Raw, in rendering fine detail and exceeds Capture One Pro in rendering smooth, artifact-free edges. The results are the most detailed, natural looking images I have seen from the 5DMKII. The crops below include the two previous best previous incarnations produced with Capture One Pro and Perfect Raw, as well as crops from Lightroom 2 and Lightroom 3 beta processed using identical settings.

Note that both the Capture One and Perfect Raw samples have been post processed to optimize detail and edge definition. Both Lightroom samples are presented without any additional post processing. To my eye, the new raw processing engine in Lightroom 3 beta renders the finest, most natural looking detail. Of course, none of these differences will be visible in any but the largest of prints. Never-the-less, the improvements in LR3 raw processing may just scratch the itch for ever more detail and obviate the need for additional software.

Raw Converter Comparison Part 2, High ISO Noise and Dynamic Range is coming soon.