Saturday, August 15, 2015

Pixel Mania: Canon’s New 5DS and 5DSr

Remember the old joke:if a little is good, more is better, and too much is just right!” In June, Canon released two new cameras that address the “just right” category. The question is, from the perspective of a landscape photographer, do I need 50 megapixels and, if so, what do I sacrifice in return? Here's my take on that subject after having played with my recently acquired 5DS over the last few weeks.

Megapixel Perspective
You’ll read that this camera is the ticket for landscape photographers who want to produce “large prints”. It’s never clear to me just how large “large” is, but regardless, there is one fundamental fact that may make it a moot point for many photographers. The human eyesight has limited resolving power, and beyond that, more resolution simply isn’t perceived. Convention is when you view any print, from a 5 x 7 to a billboard, the appropriate viewing distance is no closer than the diagonal length of the print. You can try that out yourself and see that's a reasonable assumption. Given that premise, then you’re not likely to see much, if any, improvement in resolution beyond a very good 12-megapixel camera (or there about) no matter how large the print is. 

Human Nature
Unfortunately, many of us love to scrutinize up close at the fine details of a print, even if it has nothing to do with its proper presentation. Then there are photographers (myself included) that find security in all that extra detail in the same way a squirrel feels secure with its stash of acorns: you never know when you might need it. Maybe someday you’ll make a wall mural or you just want the extra cropping space. 

So if your “big print” is 12 x 18 inches and, instead of viewing it at 22-inches, you view it at 12-inches away (and you have excellent eyesight), guess what: you'll need all those pixels from the 5DS along with good optics. So there you have it. If you want the confidence your prints can withstand scrutiny at any distance, then you’ll want all the extra pixels you can get. If you are content with a controlled “gallery” viewing environment, then your present 20 or so megapixel Nikon or Canon will do just fine.

Are Inkjet Printers Up to the Task?
Can a good inkjet printer do justice to a 50-megapixel image? I downloaded DP Review's test images for the Canon 5DSr (version without antialiasing), the Canon 5D MK3, and the Nikon D810. Without any raw or Photoshop manipulation, I sized each as a 16 x 24 inch print, which is the maximum size I can print. I did apply output sharpening using Nik’s Sharpening Pro. I then configured my trusty Epson 3800 for the highest-quality print output and focused on the extremely fine scrollwork of a foreign bank note.

And by golly, all that 50-megapixel detail does shine through! Below are scanned portions of each highly-magnified banknote (shot at ISO 800) that give a fair idea of the printed differences. (Note: scan resolution was 1800 dpi and the print paper was Epson Ultra Premium Photo Paper Luster.) The 5DSr was as expected the standout, but Nikon was a respectable second, while the MK3 blurred all the fine detail. However, it took a magnifying glass to see all that! I have corrected vision but aged eyes, and without the magnifier I couldn’t see the difference. But, younger eyes that can focus closer may spot it. Of course as the print size diminishes, that extra detail loses out to the printer’s finite resolution. But if small prints are all you care about, then you obviously don’t need a high-megapixel camera. And as for viewing on a computer monitor, a 50-megapixel camera has no advantage over any 20+ megapixel camera, even on my 5K iMac (unless, of course, you’re zooming in).

Canon MK3: Blurry, but still sharp at a proper distance.
Nikon D810: Pretty good!
Canon 5DS: Momma mia !!
“Buddy, can you spare $3,700?”
All those megapixels do come at a price—literally. The 5DS is $3,700 and the 5DSr is $200 more—not cheap! As a note for comparison, you can buy the Nikon D810 along with a 24-120mm zoom for the same price as the 5DS. Surprisingly, Canon offered a tax-free incentive sale on the 5DS less than a month after its introduction (in the Los Angeles area), which is very unusual for a new model. That made me speculate that either everyone wants the 5DSr (which didn’t have the incentive and is harder to find), or Canon misjudged the demand for ultra-high megapixel cameras.

Teeny Weeny Pixels
50 megapixels obviously bring up the concern over noise and dynamic range. I have analyzed so many raw files from the Canon 5D MK2 (my previous camera), 5D MK3, 7D MK2, 5DS, and the Nikon D810, that my head spun. Canon is often criticized for it poorer performance in noise and dynamic range compared to Sony-based sensor cameras (including the Nikon D810)—and it’s all true to some extent. I’ll admit before buying the 5DS, I seriously considered the Nikon D810. But further analyses (including the costs of buying new lenses) convinced me the 5DS was the right choice. The convincing factor was putting noise and dynamic range into proper perspective. 

For starters, I always felt my older Canon 5D MK2 was (at the time) a low-noise camera. Though I always shot at ISO 100, I hardly ever applied any noise filtering. Even when I shot at ISO 800, I only needed a little noise filtering. It wasn’t until I started doing HDR that noise became a bit problematic. But that was solved when I stopped pre-sharpening the raw file if destined for HDR (I follow the Bruce Fraser sharpening process). Furthermore, after adding just a touch of noise filtering to a 5DS test image (shot at ISO 100) it evened the noise difference with the Nikon D810 for the same image. This had to be observed at a very-high zoom level since the noise was extremely low anyway; and in this case, normally you wouldn't have bother to apply noise filtering in the first place. But the point is, afterwards I couldn’t detect any reduction in the 5DS’s resolution. Bottom line: if you always shoot your landscapes at low ISO levels (as you should), noise is simply not an issue with the 5DS—period!

Dynamic Range
As a previous large-format Velvia film user, I felt my 5D MK2 had amazing dynamic range—roughly 5-stops better! Now with my 5DS, I have verified that it has a little less noise and a bit more dynamic range in the deep blacks than the MK2, which is amazing considering it has over double the pixels. (Note, I did verify that it was ever-slightly more noisy than the MK3, which all falls in line with what Canon has stated.) In any case, from my perspective, I went from already satisfying performance to even better dynamic range and lower noise (albeit slightly), but now with tons of detail. 

Nikon D810
The Nikon D810 is a popular camera, and for good reason. Anecdotally, when viewing published photos, the Nikon D810 (and the previous D800) and the Canon 5D MK2 and MK3 seem most popular with landscape photographers. All my image analyzes did prove that the D810 does have a little less noise, even compared to the 5D MK3 at low ISOs. However, I wasn’t able to judge dynamic range to any degree. DxOMark says the Nikon D810 has 2.4-stops more dynamic range than the 5DS. However, I’m not sure what to make of DxOMark. For one, they measured a 14.8-stop dynamic range for the Nikon D810—and that was achieved with a 14-bit converter??! Given that and other considerations, I personally have a hard time swallowing DxOMark's measurements. Nevertheless, given the D810’s lower noise, and that noise and dynamic range are interrelated, it’s reasonable to assume that the D810 has an edge in dynamic range. However, I’m skeptical that the difference is that apparent in general usage (see next topic).

Dynamic Range Illusion?
It's possible that many of the anecdotal or off-handed comments made about the dynamic range differences between the Canon and Nikon are more illusional that real. I’ve discovered a noticeable difference between the way Adobe Camera Raw, and Nikon and Canon’s software render raw files. A case in point is the Nikon D810 histogram below of a somewhat wide-dynamic range image. The Camera Raw histogram shows lighter dark-tones and suppressed highlights that may lead one to think there’s a lot of headroom beyond those tonal areas. But Nikon’s raw converter instead pops the histogram to the point of clipping at both ends (see below). The Canon 5D MK3, between ACR and Canon's raw converter, showed a similar but much smaller tendency while the 5DS’s histogram was in better agreement between the two raw converters. (Note that in all cases the raw converters' tonal adjustments were zeroed out.)

Nikon D810 Histogram in Camera Raw
Nikon 810 Histogram in Nikon's Capture NX-D
So which one is right? It doesn’t matter since there isn’t any data loss, it’s just how the raw converters chooses to mess around with the image. Changing the camera profile will alter the histogram somewhat, but since part of the editing procedure is to properly distribute the histogram, it should all even out in the end.

Hands-on Experience
Image Detail
Compared to my MK2, the deep, dark-tones in the 5DS have less noise and slightly more detail. Overall, the increase in resolution is mind-boggling. I can increase the zoom level into the “electron microscope” level that goes well past the pixelated image of the MK2. But there’s good news and bad news. The good news is regardless your lenses quality (unless they're really lousy), you will see substantially more detail; but it's likely many of your lenses (especially zooms) won’t come close to doing full justice to 50 megapixels. Canon has listed L-Series lenses they recommend for the 5DS that were introduced over the last few years (none of which I have). But if you have the “cheaper” L-Series, like the popular 24-105mm and 17-40mm zooms, you will still see definite improvement. It’s the extreme edges that will benefit the least. On the other hand, if you have L-Series prime lenses, then you’ll be in hog heaven.

Image Appearance
Because of the histogram variances mentioned above, your first impression in the raw converter may be that the 5DS images are more contrasty. That may be good or bad, but it won’t matter anyway after you properly edit the image. Though I’m sticking with the Adobe Standard profile, if you want to tone down the highlights initially, try the Camera Neutral profile. I also noticed that the Auto White Balance on the 5DS produced a warmer image than the MK2. Of course that won’t matter since you can change the white balance in the raw converter to whatever suits you.  

Best New Feature
I’m fanatical about keeping the horizon level and always had used an external level. The MK3 had introduced an electronic level that was visible in the viewfinder. Now Canon made it even better on the 5DS by providing a dedicated and less obtrusive scale (at the top of the viewfinder) instead of using the focusing points. I checked it against my Kaiser shoe-mount level and they agreed as accurately as I could eyeball it.

One Annoyance
This may matter only to MK2 users since I’m not sure what the MK3 did in this regard. I always shot bracketed frames (for either exposure insurance or HDR) using the 2-second self-timer and mirror lockup option. That allowed me to avoid, in most cases, a cable release with the added insurance of no mirror shake. Plus, all three frames fired automatically. The 5DS introduced a new feature that allows you to select various shutter delays after the mirror locks up. You can still use the self-timer and mirror lockup, but in all cases the camera no longer fires the bracketed shots automatically. Instead, you have to fire them individually. In my book that’s plain stupid! However, if you skip the mirror lockup and just use the self-timer, then it fires all the frames automatically (go figure). I can live with that only because the new motor-driven mirror is unlikely to cause any problems except maybe with a super telephoto lens with razor sharp optics.

Battery Life
Based on anecdotal usage (and supported from what I’ve read in a few user reviews), the 5DS definitely burns more battery power than my MK2. My experience may be skewed somewhat since I’m doing a lot of menu surfing as I go through all the camera's features. Regardless, you’ll want backup batteries when out in the field. I have shorten my LCD display time and time-to-sleep settings. I also recommend you avoid, or at least minimize, using Live View.

Write Speed
I’ve seen a few reviewers complain about this. I bought the fastest CF and SD memory cards from SanDisk: the Extreme Pro. My 5DS is configured to record redundantly to both cards. From shutter click to LCD display, it takes almost 2-seconds versus 1.5 seconds compared to my old MK2 (using admittedly slower SanDisk Extreme III cards). Unless you’re hyper Type-A, I don’t think you'll notice much difference. The real issue is with post-processing, and of course that depends on your computer configuration. I recommend you download some 5DS raw image files and run benchmarks to make sure you're not blind sighted by agonizing process times.

Here are a few “relative’ comparisons based on my 5K iMac with 16GB memory and an i5 (4-core) Intel chip. These are based on my MK2's 20-megapixel files. The first time is the MK2 and the 5DS is the second time. All times are approximate.
  • Launch raw file from Bridge to Camera Raw: both were too short to measure
  • Transfer from Camera Raw to Photoshop: 5 sec vs. 9 sec
  • Launch three raw images from Bridge to Nik’s HDR Efex Pro (to the edit page): 42 sec vs 92 sec
  • Time to transfer from HDR Efex Pro back to Photoshop: 12 sec vs. 28 sec
  • Time to perform a Tonal Contrast midtone change in Nik’s Color Efex Pro: 8 sec vs. 10 sec.
Buy, Don’t Buy, or Wait?
This may depend on which direction you’re coming from. Most likely, anyone buying a high-end camera is already committed to that brand. But if your are starting from scratch or have little invested in legacy lenses, then let me state up front that it’s hard to argue that the Nikon D810 isn’t the top choice just based on value alone. One caveat: that opinion may change after more is learned about Sony’s new A7rII. But if you’re a dedicated Canon user and have legitimate need for more resolution that doesn't compromise noise or dynamic range, then the 5DS or 5DSr is just the ticket. 

Don’t Buy?
If you have either the 5D MK3 or older MK2 (or any APS models) and are satisfied with your 8 x 10 and occasional 11 x 14 prints, then save your money. The same applies if you do a lot of video or rely on high ISOs. The argument changes only if you do a lot of 11 x 14 or 16 x 20 prints and occasional custom poster-size prints, then it boils down to mainly if you want to spend the extra cash.

There are two good reasons to wait. First, I have noticed with Canon, Nikon, and even Sony, that their new model cameras often have teething problems. Remember the light leak issue with the 5D MK3? Sometimes it’s just a firmware upgrade. Other times you have to send it back for repair, which is a bummer for a brand new camera! So far, I haven’t experienced any problems and nothing has popped up in any reviews; but these problems can be subtle and take some time to surface. If you’re not in a rush, wait 6-months and then check user reviews for any signs of trouble.

The other reason concerns the next generation 5D MK4. Before the 5DS popped out of nowhere, I expected the successor to the MK3 would rival the Nikon D810. That now doesn’t seem likely, so based on the Canon Rumor website and my own crystal-balling, this is what I predict. The big market discriminator between the 5DS and 5D MK4 will probably be 4K video. Assuming that, pixel count may go up slightly (probably not above 28-megapixels) and actually may drop to 18-megapixels. Dynamic range and noise are likely to be better than both the 5DS and MK3. I don’t think you’ll see this camera until early 2016 at the soonest. Now take all this with a grain of salt, but if 4K video is high on your wish list, I’d wait for the MK4. 

5DS or 5DSr?
The 5DSr cancels out the antialiasing filter and I can detect a very slight increase in sharpness. But it’s so subtle, I feel the difference is insignificant. Plus, since I use zoom instead of prime lenses, I doubt I’d realize any benefit (assuming there was any real benefit to be gained). As for moirĂ© patterns with the 5DSr, judging from what B&H and Amazon reviews have said so far, that seems to be a rare problem, especially with landscape photography. To me, the decision boils down to the $200 difference. If you want the warm fuzzy of microscopically better resolution, then get it; otherwise, save your money. 

Monday, July 13, 2015

Get "Mean" with Noise

The two common ways to deal with digital noise are: shoot at a low ISO or apply noise reduction in post-processing. Traditionally when shooting landscape scenes, a tripod and low ISO went hand-in-hand and noise wasn't usually an issue, especially with low-noise cameras. Nevertheless, situations arise that require high ISOs to combat wind or curb excessive exposure times in low light. Furthermore, shooting HDR or using image enhancement plugins (for example, Nik’s Color Efex Pro), can exaggerate even minor noise to where it becomes objectionable. 

When dealing with noise during post-processing, Photoshop, Camera Raw, and Lightroom offer excellent noise reduction filters. There are also Photoshop plugins, such as Nik’s Dfine, that specialize in reducing noise. All these filters do a good job at removing noise while hardly effecting edge sharpness. The devil, though, is in the details where subtle textures and fine details may wash out. 

The “Mean” Approach
Another approach is to simply take redundant exposures and average them in Photoshop CC. In fact, some cameras do this internally, but in either case, this nondestructive approach substantially reduces noise without effecting any part of the image. The only restriction, obviously, is the scene must be completely static during the exposures. 

Averaging Procedure
  1. Shoot two to four redundant exposures (four being ideal) using, obviously, a tripod. Best to shoot at manual exposure and focus, but you probably can get away with using auto.
  2. After processing the raw files in Camera Raw (and skipping any noise reduction), from Bridge select the files and click: ToolsPhotoshopLoad Files into Photoshop Layers. You can also load the files into stacks directly within Photoshop by clicking: FileScripts → Load Files into Stack…browse and select files.
  3. In Photoshop, select all the layers and click: LayerSmart ObjectsConvert to Smart Object.
  4. After the conversion is complete, you have a single Smart Object  layer. Now click: LayerSmart ObjectsStack ModeMean. You can also select Medium which is the center value, but the results are likely to be similar.
Averaging versus Noise Filters
To see if this approach is worth the added inconvenience of shooting multiple frames, I used a dimly-lit indoor scene to compare between frame averaging, noise filters, and no filtering. The longest exposure (at ISO 100) was 2.5 sec at f/5.6. This test scene allowed me to scrutinize text and barcodes for edge sharpness and the woodgrain pattern for fine-detail rendering. I also examined the blacks and deep shadows to determine the extent of grain patterns.  

I made test shots at ISO 100, 800, and 1600. I used a Canon 5D Mark II, which is a reasonably low-noise camera, but isn't par with the latest full-frame cameras. For the ISO 800 and 1600 shots, I compared them with no noise filtering, frame averaging, and using the Dfine noise reduction plugin. I first compared two different swatch areas from the scene at 400% to gauge the relative differences. Then I printed a large section of the image at the equivalent size of a 16x24 inch print (at 234 ppi, the best my camera can do at that size) and visually assessed the image quality. 

Results at 400%
Below is the test scene. Below that are the 400% zoomed swatch samples taken from the black notebook and a section of woodgrain (see red circles). For comparison, the left image in all the samples is the ISO 100 shot without noise filtering. All middle samples are ISO 800 and ISO 1600 on the right. You can click on the images for a better view.

Test Scene: Red circles indicate the location of the 400% enlargements below.


No noise filtering

Left-to-Right: ISO 100, 800, and 1600, all with no noise reduction.
Above, you see the expected increase in grain at the higher ISOs. Though you can perceive some slight noise in the ISO 100 black patch, at this level of zoom it is insignificant enough to be considered nonexistent. 

Nik's Dfine Plugin

Left-to-Right: ISO 100 without noise reduction, ISO 800 and 1600 with Dfine plugin.
In the above black patches, the Dfine-filtered ISO 800 and 1600 samples yielded a purer black than the unfiltered ISO 100 patch. However, this came at a price you can see in the woodgrain samples. By ISO 1600, the woodgrain structure and fine detail is obliterated to a noticeable degree. 

There is some irony to this. Without divulging what to look for, a person may prefer the Dfine rendering over the virtually noiseless ISO 100 sample because the woodgrain has a more smooth and natural look. This illustrates how subjective all this can be. Nevertheless, sticking to "technically" best, elsewhere within the test scene Dfine tended to blur out other fine detail. In short, Dfine's effectiveness excels at low-frequency and gradually diminishes towards the higher frequencies (though edge sharpness wasn't noticeably affected).

Four-Frame Averaging

Left-to-Right: ISO 100 without noise reduction, ISO 800 and 1600 with 4-frame averaging.
Averaging four exposures produced the best balance between significant noise reduction and maintaining detail without degradation. Surprisingly, the black patches were all similar in appearance. Furthermore, the ISO 800 woodgrain patch rivaled the non-averaged ISO 100 patch, and the ISO 1600 was only a little worse off.

16x24 Inch Print Results
The above 400% zoomed patches are like looking at an ant under the microscope: first a scary monster, but then with the naked eye, an innocent bug. Observation made at this high a zoom level don't necessarily translate into the final print. 

I used my Epson Pro 3800 at 1440 dpi on Epson Premium Luster paper to print a large swath of the test image at an equivalent 16 x 24 inch print size. I viewed the prints from nose-length to arm-length, which the later is still technically too close to view a print this size (minimum distance is 29-inches). But bowing to human nature, I know many can't resist judging prints of any size that way. I did not, however, judge the prints using a magnifier since that would be meaningless. All test images were unedited except for the indicated noise filtering and sharpening for output using Nik's Sharpener Pro 3. 

Edge Sharpness
Resolution should suffer as the ISO increases, but all the samples, including the heavy-handed Dfine filter, were nearly equal in preserving  the resolution of small text (on the aerosol can) and UPC bar codes. Instead, image casualties were largely confined to the fine details in the woodgrain and visible noise grain in the dark shadows.

Fine Detail Rendering
ISO 100
  • Without requiring any noise filtering, the ISO 100 print is visually noise free. To validate that observation, I averaged  four frames shot at ISO 100 and compared it to the single-frame version. I couldn't detect any difference. The ISO 100 print was the gold standard I used to compare the rest.
ISO 800
  • No noise filtering: Some noise is visible when compared to the ISO 100 shot. Most noticeable was the slight breakup in the woodgrain detail. Noise was also detected in the deep shadows, but surprisingly the blacks all looked about the same.
  • Dfine noise filtering: The Dfine print is totally eradicated of any visible noise, but at the cost of rendering the woodgrain with a "creamy" look. However, as I commented before, without the ISO 100 print to compare, most (even critical) observers would likely not pickup on that flaw. 
  • Averaging frames: This was  the undisputed champ. Any detected differences between it and the ISO 100 print were more imagination that reality. 
ISO 1600
  • No noise filtering: Considerably more noise is evident and even the pure-black areas started to show slight noise. 
  • Dfine noise filter: The Dfine print, as with the ISO 800 Dfine print, was devoid of any noticeable noise. However, the "creamy look" of the woodgrain was more apparent. Even if most viewers still couldn't pickup on this defect, I think most fine-art printers would cringe at this much fine-detail loss. 
  • Averaging frames: Again the champ. Compared to the ISO 100 print, noise was only slightly more noticeable and I doubt even the most critical printer would find it objectionable. 
What about Photoshop and Camara Raw Noise Filters?
To keep this article short I excluded showing Adobe's noise filter examples. If I did, you'd see the results to be roughly between averaging and Dfine. One interesting observation was when I tried to push the Photoshop filter to same noise-reducing level as Dfine, the red graphics and red lettering on the aerosol can were interpreted as color noise and either were smeared or completely desaturated. These aren't typical image characteristic in most landscape scenes so, in practice, this probably isn't a common problem. Regardless, if you need to deal with seriously noisy images, Dfine may be the best tool of the bunch.

In-Camera Frame Averaging (updated)
To the best of my knowledge, I am only aware that certain Canon models have the ability to average multiple frames. Most cameras allow multiple exposures, but I wasn't able to determine if other brands allowed averaging the frames. The advantage to in-camera averaging is (at least in Canon's case) the operation can be performed using the raw files with a resulting raw file output. To see if there was a difference, I took an in-camera averaged image taken with my new Canon 5DS and compared it against the same image averaged in Photoshop (using the same four exposures in both cases). I examined both images unedited in Photoshop at up to a 600X zoom level and they were virtually identical.  

First a reality check. To keep things in perspective, all the preceding critical comments were more academic that meaningful, especially to the average Joe. In the real world, a great image is all that matters and nobody cares about a few scattered areas of noise grain, even if they notice it. When I was examining the 16x24 test prints, I kept asking myself if I were handed these prints without explanation and asked to find any differences, how long would it take me to see any, if I could at all! The simple fact is, if someone is noticing the noise, then the picture wasn't all that hot to begin with.

That said, I have nonetheless convinced myself that frame averaging is the best way to lower noise. From a practical standpoint, frame averaging at ISO 800 produced essentially the equivalent image quality as at ISO 100. Even at ISO 1600, the results were sufficient to satisfy most fine-art connoisseurs. The big letdown, unfortunately, is most situations requiring higher ISOs usually mean frame averaging isn't practical. You obviously can't have any subject movement and that makes it useless when increasing ISO to combat wind. So when is it useful, at least for landscape photography? Obviously if your camera is inherently noisy, this helps a lot. But I may find it useful when shooting HDR.

A case in point is when my longest exposure within a bracketed HDR series exceeds my camera's maximum shutter time of 30-seconds. This is not uncommon when shooting during the last vestige of evening light with a small aperture for greater depth-of-field. Instead of hassling with the bulb setting, I would normally bump up the ISO and deal with any noise in post-processing. Now I'll shoot between two to four consecutive series and average them without suffering any noise penalty. In fact, I may do this as a matter of course for all HDR shots. HDR software, by its nature, can at times greatly exaggerate even minuscule levels of noise. Ditto to some image enhancement software, such as Nik's Color Efex Pro. As a rule, I always strive to make my files as squeaky clean as possible, and now frame averaging is another tool to help with that.

Bonus Feature!
Here's a bonus feature with frame averaging that hasn't anything to do with noise. Did you ever long to create those dreamy water scenes taken with extremely long shutter times? Usually, you need a strong neutral density filter (e.g. ND9) to create those long exposures. But what if you don't have an ND filter? Instead, use as long an exposure time you can muster and then take repeated exposures spaced over a suitable time. Now, use the frame averaging in Photoshop CC and, voilĂ , you have a dreamy waterscape.

Wednesday, May 13, 2015

Lightroom CC's New Panorama Merge

Beside Lightroom CC’s new HDR Merge feature (see previous post below), Adobe has added a panorama merge function. And what makes this significant is that it’s better than Photoshop's panorama merge function. The reason is you can edit the stitched image in the Develop module as a raw file. In Photoshop, you edit only with the usual gamma-corrected image file. Raw files provide more editing latitude; and, as a rule, you should do as much possible of your editing with the raw converter before transitioning to Photoshop. Admittedly, that advantage may not always be apparent when dealing with a properly exposed image that doesn’t have problems in the dark tones or highlights. But if you have highlights or shadows that you need to dig out any detail from, it's better to do that in the raw converter. I’ll demonstrate that shortly with a few example images.

The panorama merge function in Lightroom is a snap to use. Adobe has a short tutorial (click here) that explains everything, including how to do a HDR panorama merge. To test out this new feature, I compared a panoramic image created in both Lightroom and Photoshop of the Queen Anne Cottage in the Los Angeles County Arboretum. I also created a HDR panorama to see how well it performed. By the way, if the image seems vaguely familiar, remember the opening dialogue of a long-ago TV series: The plane! The plane!

Routine Panorama
I started with a  two-frame panorama that was properly exposed with little or no shadow and highlight problems. For the Photoshop version, I used the Camera Raw Filter and duplicated the same settings used in the Lightroom Develop module. In this example, the raw converter's advantage in Lightroom doesn't come into play that much. Both images are essentially identical, and this is likely the case for most evenly-lit and properly exposed panoramic images. 

All the panoramas were merged using the Cylindrical layout option, which I find is best for most my panoramas. However, you'll lose a portion of the image after cropping, and that's a problem if the scene was tightly framed. In that case, you may need to use the Reposition option in Photoshop, but for some reason that wasn't included in Lightroom. The problem with the Reposition option is its propensity to create artifacts, especially if the camera was tilted when panned. Photoshop has tools to deal with some artifacts and you can click here to read more about them. The best solution, though, is always allow plenty of crop room when composing a panorama.

Lastly, I noticed that the stitched panoramic file resulted in a .dng format (Adobe's open-standard raw file format) and not your camera's proprietary raw format. Not an issue for me since I convert all my Canon raw files to .dng anyway.

Lightroom CC Panorama Merge
Excellent results and no stitching artifacts.
Photoshop 2014 Panorama Merge
Both Lightroom and Photoshop are essentially identical.

Highlight and Shadow Recovery
In this example, I used a 3-frame panoramic shot of the same scene. One series was shot at +1.33EV overexposure and another at -1.33EV underexposure. This intentionally created some problematic highlights and shadows that I then attempted to extract detail from. The image below shows which highlights (blue-circled cottage roof) and shadows (red-circled palm trees) I worked on.

Highlight/Shadow Recovery
I'll demonstrate the recovery difference between Lightroom and Photoshop within the blue-circled
highlight area (cottage roof) and red-circled shadow area (between the palm tree trunks).
Highlight Recovery
In this example, the Develop module and Camera Raw Filter settings were identical. The Lightroom version has better overall contrast and detail in the highlights, plus better texture in the white plastic tarp. To read more why editing in raw is better, see my post on Exposing-to-the-Right. There, I discuss the advantages of a linear-data file (raw) versus a gamma-corrected file (Photoshop). There are other reasons too, such as a special highlight recovery algorithm in Adobe's raw converter.

Lightroom Highlight Recovery
Better overall contrast and detail, especially in the white tarp's texture.
Photoshop Highlight Recovery
Maybe a bit more fiddling in the Camera Raw Filter could even things 
up a bit more, but fundamentally, Photoshop just has less data to work with.
Shadow Recovery
At the dark end of the histogram, the differences are more noticeable. In the Photoshop version, I had to apply additional shadow recovery to the Camera Raw Filter in an attempt to even the difference. But in spite of the additional tweaking, the Lightroom version is still noticeably superior at rendering the shadow detail while the same detail in the Photoshop version is blocked up. In addition, the overall Photoshop image (including the shadow detail) was a little too contrasty. 

Lightroom's Shadow Recovery
Lightroom does a really a great job! Additional editing
could perk up the contrast and saturation a bit more
and still retain good shadow detail.
Photoshop's Shadow Recovery
 Blocked-up shadow detail and too much contrast,

HDR Panoramas
I didn't bother to compare the HDR panorama performance between Lightroom and Photoshop since Photoshop's 32-mode has gone squirrely in their latest 2014 version (at least on my computer). Nevertheless, I tested how well it worked in Lightroom, and the results were excellent. 

Now there are two ways to process a HDR panorama shot. You can stitch all the frames for each exposure, then combined them in HDR. Or, first combine the exposures for each frame segment in HDR and then stitch them into a panorama. However, only the latter will work in Lightroom, and for that matter, only in Photoshop and Photomatix 5 as well. The reason is when you stitch each exposure frame set, each set results in a slightly different size (by only a few pixels). That stops Lightroom and Photoshop dead in their tracks, and while Photomatix will merge the frames, it generates weird artifacts. Only HDR Efex Pro 2 seems to cope with the slightly different image sizes without a hiccup.

The reason I bring this up is because, intuitively, I prefer to combine the stitched panoramas in HDR. One reason is I worry that first combining each segment into HDR before stitching may produce slightly uneven results between the panoramic segments, especially in luminance. Any differences in luminance will become apparent in the sky after they're stitched. Of course, as already mentioned before, you can repair the artifacts in Photoshop, but who needs the hassle. But the biggest reason is I want the ability to fully edit the whole panorama in 32-bit mode. Yes, I could edit each segment in 32-big mode, but that's both awkward and may increase the chances of uneven frames. 

Nevertheless, simply merging each segment into HDR (without, of course, any additional editing in 32-mode) and stitching them into a panorama produced an excellent image, at least for this example. And as a consolation prize for losing the 32-bit editing capability, the stitched file at least remains a linear file. However, this image wasn't particularly challenging and a high-contrast sunrise or sunset scene may beg for 32-bit editing capability. Then the only recourse I can see is to stitch each exposure series in either Lightroom or Photoshop (probably doesn't matter which) and use HDR Efex Pro 2 to edit the entire image.

Lightroom HDR Panorama
Results were great with no artifacts (HDR or stitching) and even sky illumination. Highlight
and shadow details are excellent. A little more editing will brighten up the image even more.
The preceding discussion was as much about Lightroom's new panorama merge as why editing in raw is important. But I needed to emphasize the latter to illustrate how nifty the new Lightroom feature is. Does this replace Photoshop's panorama merge (aka, Photomerge)? I'd say pretty much so. I'm hard-pressed to think why you would ever want to circumvent editing in raw. If Lightroom's merge did create any stitching artifacts, you can still fix them in Photoshop after you completed the bulk of editing in Lightroom. The only reason I can see to use Photoshop is for the Reposition layout option; but proper composing should avoid that need.

Wednesday, May 6, 2015

Lightroom CC’s New HDR Merge

Lightroom CC has a new HDR feature that merges multiple exposures for 32-bit tone mapping within the Develop module. This mimics the existing 32-bit mode featured in Photoshop that merges exposures using HDR Pro, but tone maps in Camera Raw rather than in HDR Pro. (Click this link to see my previous post on Photoshop CS6's 32-bit mode.) Processing HDR files in Lightroom CC is very simple and rather my explaining it, just go to this link for a short Adobe tutorial. To test out Lightroom CC's new HDR mode, I compared it to Photoshop's 32-bit mode, HDR Efex Pro 2, and Photomatix Pro 5. (Note: Also new in Lightroom CC is a very promising panorama merge feature that I’ll report on later.)

Say It Ain't So Photoshop!
My expectations were that Lightroom and Photoshop would perform identically since, after all, they’re essentially the same software. Well, that expectation was sorta true. Lightroom seemed to perform similarly to the results I observed when I originally evaluated Photoshop C6’s new 32-bit mode. But something happened on the road to bigger-and-better. My evaluation this time was based on Photoshop CC 2014 (v2.2) — and now it's awful!  The biggest issue is the rendering of dark tones, but the rest of the image is also mediocre. I did find one similar complaint on the Internet. I don’t know if it’s a Mac-only problem, something unique to my configuration, or some boneheaded error on my part (anything is possible); but as it stands now, it’s essentially unusable.

Update: After upgrading to the new Photoshop 2015, the above problem vanished. Photoshop 2015 and Lightroom CC now appear equivalent in performance, as you would expect. Any comments hereon on Lightroom CC should apply equally to Photoshop 2015's 32-bit mode.

I'll show two example landscape scenes processed in Lightroom, Photoshop, and HDR Efex Pro (and Photomatix 5 on the second scene). First, I'll show how each image initially emerged from the HDR merge process. For HDR Efex, that means how the images appeared right out of the oven before any additional 32-bit editing. For Lightroom and Photoshop, however, it was necessary to make major adjustments to all the exposure controls to get the histogram into a decent starting point. This is normal and shouldn’t be construed as a performance flaw. Next, I’ll show the results after additional 32-bit editing and final touchups in Photoshop with a little help from Color Efex Pro. Note: I didn't spend much time on color balance, so ignore any color differences which aren't necessarily a characteristic of any particular process.

Scene One: High-Contrast Daylight Scene in Bryce Canyon National Park
Lightroom CC
The initial HDR merge in Lightroom produced acceptable results, but it took radical increases in all the exposure sliders to get to a decent starting point. Though you have a workable image, the extreme exposure adjustments leave less editing latitude in the highlights and shadows during subsequent 32-bit editing. Nevertheless, the final image rivals HDR Efex Pro, but lacks the crisp contrast and finer detail in the highlights and shadows. Image quality is often more subjective than technical, so you may still prefer the Lightroom rendering over HDR Efex Pro. However, I feel HDR Efex Pro has the edge.

Lightroom Raw Merge: All the image detail is there, but it
took pushing the exposure controls nearly to their stops. 
Lightroom After Final Editing: Satisfying results
with a natural appearance that adequately captured
detail in the highlights and shadows.
Photoshop 2014
In a word: yuck! After merging in HDR Pro, the image is oversaturated, too dark, and simply garish. Previously in Photoshop CS6, the results would have been similar to Lightroom's. I had to throw everything (including the kitchen sink) at it in both Camera Raw and Photoshop to eventually eke out a presentable image.

Photoshop CC 2014 Raw Merge: Too dark, 
oversaturated, and poor detail in the shadows.
Photoshop CC 2014 After Final Editing:
Better than nothing? Yeah, but not by much.
HDR Efex Pro 2
At first blush the HDR Efex Pro raw output seems less impressive than the Lightroom version; except, the exact opposite is true. All the highlights and shadows are well defined within the histogram, which is what you need to have maximum editing latitude in those tonal areas. The final image needed the least editing gymnastics and produced a nicely toned and reasonably nature appearance.
HDR Efex Pro 2 Raw Merge: The champ at
capturing all the highlights and dark tones. 
HDR Efex Pro After Final Editing: Great results with plenty of
editing latitude to reshape it to whatever suits your taste.
Scene Two: Sunrise at Joshua Tree National Park
Lightroom CC
Again, the merged image gives you a decent starting point. And this time, Lightroom really gives HDR Efex Pro a run for its money in the final image. HDR Efex Pro still has the advantage in better detail rendering and midtone contrast, but loses in two respects. A systematic problem with HDR Efex Pro has always been two irritating flaws. The first is emphasizing noise, and the second is halo artifacts along high-contrast edges. Both are correctable in Photoshop, but that's an unwelcomed nuisance. All the processes are susceptible to this to some degree, but HDR Efex Pro is most prone. So, would I choose  Lightroom's version over HDR Efex Pro? Not yet—keep reading.

Lightroom Raw Merge: Ditto to previous example.
Also, very low in noise and halo artifacts.
Lightroom After Final Editing: Excellent
results with a natural "traditional photo look".
Photoshop 2014
There's no point in beating a dead horse. The results are just as disappointing as in the Bryce Canyon photo.
Photoshop CC 2014 Raw Merge: Badly blocked up shadows.
Photoshop CC 2014 After Final Editing: In a word: unacceptable.
HDR Efex Pro 2
The merged results are excellent with plenty of editing latitude. The final results also don't disappoint (except for slightly more noise and halo artifacts, which you can observe between the mountain ridge and sky). Probably I would still pick this photo over Lightroom after fixing the artifacts and working the sky a little better (also note that this image needs some repair for lens flare that I didn't bother with in any of these examples). But in any case, I would be very satisfied with either versions. 

HDR Efex Pro 2 Raw Merge: Repeat performance.
HDR Efex Pro 2 After Final Editing: Some annoying halos
and slightly more noise (all which can be fixed in Photoshop)

Photomatix 5
To spice up the comparison, I included a final image done in Photomatix 5 using the new Contrast Optimizer. I went back and forth on this, but concluded that Photomatix edged out the other two, but only for this particular image. The results are similar to HDR Efex Pro, but without the slight increase in noise and halo artifacts. Admittedly, I could go back and re-tweak HDR Efex Pro; but Photomatix quickly produced an excellent result for this particular image with little effort, so I'll give it the crown mainly for that reason.

Photomatix 5 After Final Editing: Unless I go back and fiddle
more with the HDR Efex Pro example, I'd say this one is the winner.
No (or barely) halo problems or noise increase.
If I had to describe the basic character of each, I would say that HDR Efex Pro excels in rendering detail in highlights and shadows, has snappy midtone contrast, and has powerful editing tools. The better rending does come at the cost of a very slight "HDR look" (i.e, slightly unnatural or surrealistic) and, of course, the already mentioned noise and halo artifacts. Lightroom has a more soft and natural appearance—more towards a traditional photo look. If that more subdued appearance is better or not, I'll leave that to your personal taste. Also, the powerful editing tools in the Develop module (or Camera Raw once Photoshop gets its act together) are a big plus, but that's slightly diminished by having to push many of the exposure adjusts to their limit. Photomatix 5 (specifically Contrast Optimizer) is roughly in-between, though a bit closer to HDR Efex Pro. It's editing ability is the most limited, but at least it's also the simplest. In spite of that, it can still bang out a stunning image. For more information on Photomatix 5 (specifically Contrast Optimizer), click this link to a previous post I wrote comparing Photomatix 5 and HDR Efex Pro 2.

If you're interested in more information, please visit my posts on Photoshop's 32-bit mode and Photomatix 5. I believe the comments and conclusions I made there are still relevant. However, for the Photoshop 32-bit mode article, substitute the name "Lightroom CC" for "Photoshop". Given how poorly Photoshop 2014's HDR Pro performed, I'm sure it will eventually be fixed. But for now, it doesn't matter anyway. If you are using CS6, then you're not affected. If you're on the Adobe subscription plan, then you have Lightroom. In either case, you have a viable HDR processor that should satisfy a good portion (and maybe most) of your HDR needs. 

But if you are a heavy user of HDR, I still recommend that you also have either HDR Efex Pro 2 or Photomatix 5. In fact, I strongly recommend you have both. In my experience, there can be erratic behavior in file merging in any of the processes. Even HDR Efex Pro, which is my primary choice, has on occasion gone inexplicably bonkers on an image. Remember, you can't preview your HDR images out in the field. Unless you shot a backup frame using just a graduated neutral density filter (something I always recommend), you may be at risk of losing a once-in-a-lifetime shot unless you have more than one way to process your HDR images.