DP1 Shootout Pt. 2 - Dynamic Range Comparison with Canon 5D and Olympus E-420

Much of the discussion regarding Sigma's cameras, with their Foveon sensors, has focused on the low megapixel count and "high per pixel sharpness" of the native output files. Yet at the print sizes most of us tend to use, neither of these characteristics, one of which tends to balance the other, dominates image quality. On the other hand, the dynamic range of the output and the range of tones within that range have a significant impact on image quality at any print size.

The dynamic range (DR) capability of a camera depends on the sensor technology used (see more discussion here and here); but in general, for a given type of sensor technology, DR will correlate with sensel (sensor "pixel") size. The DP1 has a very large sensel area. Given this and anecdotal reports from Sigma DSLR users about the DR of those cameras, I was optimistic about the DP1 DR.

Dynamic range test results can often be confusing. Two reputable sites will use the same method, such as Imatest, and come up with significantly different results. There are rarely practical examples shown to demonstrate the detected differences. Furthermore, dynamic range testing is often done on in-camera JPEGs, so the information gained doesn't reflect the entire sensor capability and is of limited practical use to me as a primarily RAW shooter.

My method is pretty straightforward. I take a single photo of a high DR scene, one which exceeds the DR capability of the camera sensor. I then "push" the image to +2EV and examine the shadow detail. Next, I push the same image to -2EV and look at the highlight information. The ability of the camera to simultaneously capture both the shadow and highlight detail from a challenging scene can thus be determined in a practical manner. It is critical that a single capture be used to look at both shadow and highlight latitude. With any camera, it would be easy to demonstrate good highlight detail from one capture and good shadow detail from another, since ambient lighting conditions and camera exposure settings can be vary between two captures.

I chose to compare the DP1 DR to that of the Canon 5D and E-420 since those are the most compact digital representatives of the 35mm and Four Thirds formats, respectively. The 5D was used with the Canon EF 28mm f/2.8 lens, and the Olympus with the ZD 14-42mm f/3.5-5.6 zoom at 14mm. The Canon image was taken at f/11 and 1/100s. The Olympus and Sigma images were at f/5.6 (for a similar depth of field to the Canon) and 1/400s (for an equal exposure). The three captures (one from each camera) occurred within a minute of one another, and the lighting conditions were not appreciably changing. All were at ISO 100. EXIF information is not embedded in the crops, but the RAW files are available for download at the end of the post. For details regarding processing workflow, please refer to the shootout introduction.

First, let's look at the highlight information. The specific area of examination is depicted by the yellow box in the following resized DP1 photo:



All crops shown are 100% without resizing. Here is the Canon 5D example, with the baseline conversion on the left and -2EV on the right:



As you can see, the tonality in the sky and car windshield was completely lost in the baseline capture and partially restored at -2EV. Likewise, a number of the small branches were restored.

Next we'll take a look at the E-420 performance in this area:



The baseline crop from the E-420 is similar to that of the 5D. However, the E-420 does not gain as much highlight detail by comparison at -2EV.

Finally, here is the DP1 sample:



Unlike the 5D and E-420, the DP1 has managed to retain some of the variation in the sky and car windshield, as well as the small branch definition, even in the baseline capture. At -2EV, there some further gain.

Looking at these results for highlight latitude, it is tempting to speculate that the DP1 was underexposed compared with the other cameras. However, as the EXIF in the RAW files will testify, the exposure settings were equivalent for the three cameras. Is it possible that in the moments between captures the ambient light dropped just in time for the DP1 capture? That is possible, though no such change was apparent. One way to get at this possibility is to see how the DP1 did from a shadow standpoint in the very same capture. One would expect that if a lower exposure is the exaplanation for the strong retention of highlight detail, that shadow detail would be compromised.

For shadow detail, I examined the two regions depicted in the following image:



Here is the Canon 5D examples, with the baseline conversion on the left and +2EV on the right in each of the two regions examined:



The additional apparent shadow detail at the +2EV setting comes, predictably, at the expense of greater noise.

Next we'll take a look at the E-420 performance in this area:



The comments from the 5D crop apply here as well.

Finally, here is the DP1 sample:



The pushed DP1 shadows were, to me, surprisingly clean. Clean enough to make me wonder whether some noise reduction is being applied by SPP during RAW conversion.

Based on this comparison I would conclude that DP1 exposure latitude at the base ISO is similar to that of the 5D and exceeds that of the E-420. Other comparisons I have done have been consistent with this observation.

You can download the RAW files for the three images used in this comparison by clicking the links below.



In Part 3 of the shootout, we'll take a look at the overall detail captured using the DP1 compared with the Canon 5D and Olympus E-420. Part 4 will compare the low light, high ISO performance of the same three cameras in addition to the current compact camera high ISO champ, the Fuji F31. The results may surprise you!

Addendum:
Sean Reid of Reid Reviews reminded me to discuss the impact of overall lens contrast on effective dynamic range. A lower contrast lens will tend to lighten the tones in the shadows (move them away from pure black) and, in doing so, may provide separation between shadow detail and the noise floor. The noise floor is the level at which the detail in deep shadows is lost among the file's digital noise. The effective dynamic range that a digital camera can record, therefore, is noticeably affected by lens contrast.

To truly isolate the DR ability of the cameras being compared, one would have to use the same lens on each camera. Clearly this is not possible with the three cameras being compared. Thus it is best to think of the results shown above as reflecting the particular camera-lens combinations used.

Posted by Amin

4 comments:

Anonymous said... April 20, 2008 at 2:49 PM  

Thanks for taking the time to do a nice comparison. Very interesting stuff. :-)

I hope a later installment will cover overall sharpness of the cameras/lenses, and distortion.

Eventually even a little handling/size comparison would be cool IMHO.

The reason I am requesting these types of things, is that I have an E-410 and my buddy has a 5D that we use for photographic work. Having handled those lovely cameras I would like to know how the rather unique and cool DP1 "feels" in comparison to them.

Keep up the great work!

Amin said... April 20, 2008 at 4:08 PM  

Yes, I do plan to do those comparisons. It will take me a little while to get to everything, but they are coming. Thanks for taking the time to look and for the kind comments. Regards, Amin

Andrew said... April 21, 2008 at 8:19 PM  

Thanks for the comparison. I opened up the 5D file in ACR 4.4.1, and I noticed that it does a much better job in the +2.0 scenario. Even with color noise reduction set to 0 (everything else 0'd as well), there is much less noise than the result that was posted.

It'll be great once there is a RAW convertor that will support all 3 of these cameras.

Amin said... April 22, 2008 at 5:51 AM  

Thanks for your comment Andrew. Actually my perception that ACR (since v4.1) applies NR even when NR is set to "0" is the reason I didn't use it for this comparison. However, you are absolutely correct that ACR or C1 give a better showing for the 5D at +2EV, and I can't rule out the possibility that Master (doubtful) or SPP (? likely) are doing some NR here. I'd love to be able to process all three with Raw Developer, because that application tends to show all the warts, preserving as much fine detail as possible without concern for noise.

Post a Comment

Recent Posts

 
Copyright 2007 | Andreas08v2 by GeckoandFly and TemplatesForYou | Design by Andreas Viklund
TFY Burajiru