HDR vs. ACR

[John T]

With this posting I would like to relay my observations about post processing methods for images that contain high dynamic range. To navigate these waters, I started with the advice relayed in the following two books:

1) “A World in HDR” by Trey Ratcliff
2) “Practical HDR: A Complete Guide to Creating High Dynamic Range Images with Your Digital SLR” by David Nightingale

As mentioned in my last posting (Evolution of a Newbie), I experienced some difficulties in getting photorealistic images. From what I have gleaned from other postings on this and other sites, these frustrations are not uncommon to new person taking up the HDR image processing.

In response to my last posting, Colin Southern made an excellent suggestion that I read another source as listed below:

3) “Real World Camera Raw with Adobe Photoshop CS5” by Jeff Schewe, Bruce Fraser

I have found Collin to be an excellent source of information and on this basis purchased a copy of the book. I have jumped into this source and found it to provide an alternative perspective for post processing of images obtained with high dynamic range. It should be noted that this book covers themes related to Adobe Camera Raw (ACR) and is directed toward the professional photographer (which I am not, Collin is as you may note from his web gallery). The book does, however, also provide general information that can be used to generate alternative strategies to process single images. From what I can tell from my initial forays into ACR, the approach for processing raw images can compliment more traditional HDR and tone mapping steps.

With this posting I would like to share my “semi-scientific study” of the various HDR methods alongside what I have done with ACR.

The images I have chosen for the study are taken of a pathway on the island of Capri, Italy during sunrise on the south side of the island, May 2010. It is dangerous to make too many conclusions from a single image sequence, but his is a place to start.
The images were taken using standard methods for image capture based on Digital SLR (i.e. Cannon Ti1) using tripod and bracketed at 0, 2, -2 EV. An analysis of the images reveals that the ideally exposed image (EV0) has the highlights clipped in the sky, but the under exposed image (EV-2) brings this into check. The over exposed image (EV2) brings the shadows better into play. The sequence seems to fit well the ideal situation where HDR should provide an advantage over a single image exposure.



For this study I used three methods to generate HDR-tone mapped images based on:

1) Photomatix
2) FDR tools
3) Photoshop CS5 HDR

At this stage in the game, I have no particular favorite HDR software. I do see from my reading that the Photomatix seems to win the popularity contest. In any event, I have spent about the same amount of time for each of these packages and wanted to see here if there was any clear advantage relative to processing the single image.










From my execution of the HDR and tone mapping, there are comparable results obtained with the three packages. Each of the images has introduced noise which I have not attempted to resolve in this test. You will note also a lack of realistic contrast that is typical for the HDR approach. The most significant flaw of the way I have processed the images is that the sky region if left looking very muted. Again, this seems to be typical of the experience of others.

It is probably not fair for me to comment on the relative attributes of each approach as there are many permutations for each method that need to be explored. With a large grain of salt, here are my initial conclusions:

A) Phomatix seems to generate the most hyper-real image (the same conclusion is made in the HDR books listed above)
B) PFR Tools seems to generate the most photorealistic image (again, the same conclusion is made in the HDR books listed above)
C) Photoshop CS5 HDR seems to generate the dullest image (the new version of Photoshop HDR is not covered in these books)

It should be noted that the above two books that advocate HDR suggest that post image processing is necessary. Since this is not yet done here, the full potential of the images has yet to be derived. The post HDR processing with be explored below.
Next in sequence, I used the tool of Adobe Camera Raw as advocated by the third reference above. After making some adjustments in ACR, the following image was generated. Note that to complete this single image conversion, I used the most underexposed image of the sequence (EV=-2).



For the ACR processing of this dark image I was surprised to find the “Fill Light” slider does such a good job of recovering the details in the shadows. The ACR interface has some tools to reduce noise and this was employed as part of the image conversion. The ACR image looks more realistic, although some additional improvement was desired.

For the next stage of development, I moved the images into Photoshop CS5 and begun to play with layers, levels, and filters.

Next in sequence I used Topaz Adjust 4 in an effort to pop the colors a bit more. This suggestion is made in the book by Trey Ratcliff. Topaz software has a number of useful presets that can be used to quickly explore new effects. These can be used as starting points for further customization. With this pathway in mind, I began with the “Exposure Color Stretch” preset of Topaz Adjust and then toned down the effect a bit. The next image is obtained using this approach. The colors are popped, perhaps too much?



In a similar way, I used a second Topaz preset named “Spicify” to secure the next image. This image too has the colors popped.



Since I liked some of the attributes of the Spicify filter, I then went back to the Photoshop HDR image and augmented its treatment with a similarly applied Spicify filter. This doubly processed image is provided below. My assessment of this combination method is that it seems to still have that over processed HDR look.



Last but not least, I used the ACR image to bring into Photoshop CS5 and then used levels, layers and auto contrast settings to process the image. The final image here provided is obtained with this approach. This too produces decent results and yields a photorealistic image endpoint.




INPUT REQUESTED

I have attempted through this study to secure some personal experience that I hope will help me and others to process images better in the future. I solicit input from the experts in the following areas:
1) It would help to secure comments on the quality of the various processed images. Please specify which one of the images is your favorite and why. Have I popped the colors too much in certain cases.
2) Please take a look at the histograms for the source images and let me know if this image sequence is well suited (or not) for HDR.
3) Please suggest other major factors that should be considered in the processing of these images.

Thanks gain to all CIC members for your helpful suggestions.

[arith]

FDR tools is the one for me; Photomatix looks like a Topaz Large Detail or extreme in an extreme sense. The others lack shadow detail and are dull but looking at your originals it should be possible to do something with it.

[Eric M]

I prefer your straight ACR image, though it could still use a bit of edit work. The HDR program tonemaps don't do much for me unless I am going for an artistic product. That isn't the case most of the time.

Several years ago I had the HDR bug. Not long after I got into it I began shooting a project that demanded accurate color reproduction and switched to "tonemapping" RAW files with ACR. The results are much more accurate color. Of course, they aren't true HDR images either.

The secret to using ACR effectively (for me anyway) is to push the highlights almost to the point of clipping and recover them in ACR. This helps prevent shadow noise by reducing the need to boost the shadows. Depending on the scene, I also am a firm believer in using wireless flash units to control contrast and stay in a dynamic range a single RAW file can capture.

As an example, the image seen here was done using ACR and wireless flash. I shoot historic industrial subjects. This is the interior of a scale house at a defunct coke plant. Hopefully you can't tell I used two speedlights and hopefully the color is photorealistic. To my eye, it looks like what I remember seeing thru the viewfinder.

[John T]

Arith,

Thanks for the comments.

Can you explain to me what features of the Phtomatix version you prefer over the other versions.

John

[John T]

Eric,

Thanks for the comments.

Can you provide me some details on why you prefer the straight ACR image?

Also, what additional edits would you make to this image.

I have taken note of your comment about pushing the limits of exposure and the attempting to recover in ACR. In this case, I had a three exposure sequence, but chose to process the least exposed image because I wanted to have best chance to capture the sky in its richest colors. In my limited experience, I have found it possible with noise reduction software like Topaz Denoise to eliminate the noise from the shadow area. This in some cases comes at the expense of some details in the shadows.

Can you elaborate on this type of trade off?

Thanks again, John

PS. Like your image of the industrial area. The colors appear realistic.

[Colin Southern]

Hi John,

Looking at your 3 examples, I'm seeing a consistent issue: local contrast. In the first two examples the software has attempted to boost local contrast, but (as is pretty typical of these programs) has got it pretty wrong in some parts, resulting in the typical image that's similar to what one gets from too agressive a use of the clarity slider in ACR, which is in essence similar to high radius sharpening. On the other hand, the CS5 HDR attempt hasn't made any local contrast correction -- it's basically just compressed the entire dynamic range into something that'll display on our screens, which doesn't look very nice either.

For these images you have two distinct light zones (the sky, and the rest) -- personally, I'd be inclined to simply stack one above the other and then paint a transition mask so that only the good bits are revealed. I suspect that that'll take care of things nicely, and it of course means that the local contract is both automatically taken care of, and photorealistic - a win/win Happy to create you a PST file showing you what I mean if you'd like to flick me the RAW captures.

PS: Thanks Eric - you saved me a lot of typing when you said just about everything I was going to say - I agree 100%

[Eric M]

Eric,

Thanks for the comments.

Can you provide me some details on why you prefer the straight ACR image?

Also, what additional edits would you make to this image.

I have taken note of your comment about pushing the limits of exposure and the attempting to recover in ACR. In this case, I had a three exposure sequence, but chose to process the least exposed image because I wanted to have best chance to capture the sky in its richest colors. In my limited experience, I have found it possible with noise reduction software like Topaz Denoise to eliminate the noise from the shadow area. This in some cases comes at the expense of some details in the shadows.

Can you elaborate on this type of trade off?

John,

The ACR image seems to have more natural color and less of the HDR "cartoon" look to it. As for additional edits, the sky looks a little flat in relation to the landscape. The hills look like they might need to be warmed up a touch (even though they are blue from being shaded which is normal). This is one of those extreme images that could require some time to really get right. Somehow it just doesn't quite pop yet.

As for your comments on boosting the shadows in ACR and dealing with any noise that occurs, that is certainly one way to go. I'd rather not deal with the noise if possible and another reason I push the highlights is that nature usually provides a bright sky in relation to the forground in many scenes. Images like yours (backlit) can have an almost neutral of even reverse contrast if they aren't edited carefully. That provides much of the "fake" look characteristic of many HDR images.

It is neat to discuss this subject as I usually just shoot from the hip and edit away until some anchor in my mind says "that looks right". Stopping to think about the reasons why certain edits look realistic and others do not is interesting.

[Colin Southern]

John,

For what it's worth (and following on from what Eric said), I come up against the dynamic range limits all the time - and I've found that the best way to handle it is to have highlight alert and histogram display switched on on the camera ... if I get one or two blinkies in the sky then I know I'm pretty close to sensor saturation; if I don't get ANY blinkies then I'll take a look at the histogram and see how under-exposed I am, and adjust accordingly.

Looking at your -2EV image histogram it looks as if there was probably another stop available - if so then by under-exposing 1 shot you've effectively thrown away a full 1/2 of the information you could have captured.

In reality for shots that push the limits you end up throwing an exposure based on a middle gray out the window; it's all about exposing to the right, but not blowing a significant amount of highlight ... which will give you the cleanest possible shadow detail when you reveal the shadow detail with ACR.

Keep in mind too that ALL images need post-processing after the HDR compression process.

[timo2]

John, the results you have achieved are severally hampered by the fact that the initial exposures are incorrect. The 0EV one is under underexposed by more than 2 stops; if you look at the +2EV frame, you can see that even there, it is substantially under exposed.

This is not a good basis to attempt to evaluate HDR software.

[Colin Southern]

if you look at the +2EV frame, you can see that even there, it is substantially under exposed.

Hi Tim,

I'd be interested to learn how you came to that conclusion; The histogram is spiked at the end and the sky detail is totally washed out.

[timo2]

Colin, you have only partially quoted me. The sentence after the semicolon refers to the previous one. In that, I was referring to the fact that the 0EV one is substantially underexposed and I mentioned the +EV frame as a point of reference to illustrate my point; possibly badly, sorry. I’ll try and elaborate

In the +2EV frame, the spike at the end of the histogram is obviously the sky but that isn’t what the + exposures are about and so the sky histogram spike is irrelevant.

My point was that the main hump of the histogram in the +2EV frame, ie the area of the picture excluding the sky, the part of the picture that +EV exposures should be exposed for – the shadows - would have been further to the right if the 0EV frame was correctly exposed.

[Colin Southern]

Hi Tim,

I see what you mean now - sorry, must have got "lost in translation" somewhere.

Personally I still think that if a single exposure was taken right up to the max the rest of the detail would probably reveal without too much noise using just ACR.

[timo2]

No worries Colin, and you may well be right. My point was: to realistically evaluate HDR software, it (they) must be fed with correct exposures otherwise results will most likely be disappointing and lead to erroneous conclusions.

[John T]

Hi John,

Looking at your 3 examples, I'm seeing a consistent issue: local contrast. In the first two examples the software has attempted to boost local contrast, but (as is pretty typical of these programs) has got it pretty wrong in some parts, resulting in the typical image that's similar to what one gets from too agressive a use of the clarity slider in ACR, which is in essence similar to high radius sharpening. On the other hand, the CS5 HDR attempt hasn't made any local contrast correction -- it's basically just compressed the entire dynamic range into something that'll display on our screens, which doesn't look very nice either.

For these images you have two distinct light zones (the sky, and the rest) -- personally, I'd be inclined to simply stack one above the other and then paint a transition mask so that only the good bits are revealed. I suspect that that'll take care of things nicely, and it of course means that the local contract is both automatically taken care of, and photorealistic - a win/win Happy to create you a PST file showing you what I mean if you'd like to flick me the RAW captures.

PS: Thanks Eric - you saved me a lot of typing when you said just about everything I was going to say - I agree 100%

Colin,

Thanks for the suggestion. I have sent you the RAW files now as per your suggestion. Look forward to seeing your next ideas.

Note that I completed the ACR adjustments using the most underexposed image. My intent here was to keep as much color as possible to create a realistic looking sky. Up to this juncture, this has been my most challenging task. I have been able to cope with the noise using tools like Topaz Denoise.

A couple of followup questions:

1) Would you have chosen a different image with longer exposure to complete the ACR?

2) What is your take on management of noise with tools like Topaz Denoise or do you use Photoshop directly for this task?

Thanks again, John

[John T]

John,

The ACR image seems to have more natural color and less of the HDR "cartoon" look to it. As for additional edits, the sky looks a little flat in relation to the landscape. The hills look like they might need to be warmed up a touch (even though they are blue from being shaded which is normal). This is one of those extreme images that could require some time to really get right. Somehow it just doesn't quite pop yet.

As for your comments on boosting the shadows in ACR and dealing with any noise that occurs, that is certainly one way to go. I'd rather not deal with the noise if possible and another reason I push the highlights is that nature usually provides a bright sky in relation to the forground in many scenes. Images like yours (backlit) can have an almost neutral of even reverse contrast if they aren't edited carefully. That provides much of the "fake" look characteristic of many HDR images.

It is neat to discuss this subject as I usually just shoot from the hip and edit away until some anchor in my mind says "that looks right". Stopping to think about the reasons why certain edits look realistic and others do not is interesting.

Eric,

Thanks for the thoughtful suggestions.

You and Colin have made basically the same suggestion about pushing the limits of the highlights and working to recover back. I can take a step in this direction by starting with either the EV0 or EV2 images from the AEB sequence. I was a bit timid about such an approach as I was worried about getting decent looking sky. Would you suggest going this route?

Thanks, John

[John T]

John,

The ACR image seems to have more natural color and less of the HDR "cartoon" look to it. As for additional edits, the sky looks a little flat in relation to the landscape. The hills look like they might need to be warmed up a touch (even though they are blue from being shaded which is normal). This is one of those extreme images that could require some time to really get right. Somehow it just doesn't quite pop yet.

As for your comments on boosting the shadows in ACR and dealing with any noise that occurs, that is certainly one way to go. I'd rather not deal with the noise if possible and another reason I push the highlights is that nature usually provides a bright sky in relation to the forground in many scenes. Images like yours (backlit) can have an almost neutral of even reverse contrast if they aren't edited carefully. That provides much of the "fake" look characteristic of many HDR images.

It is neat to discuss this subject as I usually just shoot from the hip and edit away until some anchor in my mind says "that looks right". Stopping to think about the reasons why certain edits look realistic and others do not is interesting.

Eric,

Thanks for the explanation. This is helpful.

What are your thoughts on my use of the Topaz filters. I presume I have gone too far. Any suggestions on color popping without going too far would be appreciated.

Best, John

John

[John T]

John,

For what it's worth (and following on from what Eric said), I come up against the dynamic range limits all the time - and I've found that the best way to handle it is to have highlight alert and histogram display switched on on the camera ... if I get one or two blinkies in the sky then I know I'm pretty close to sensor saturation; if I don't get ANY blinkies then I'll take a look at the histogram and see how under-exposed I am, and adjust accordingly.

Looking at your -2EV image histogram it looks as if there was probably another stop available - if so then by under-exposing 1 shot you've effectively thrown away a full 1/2 of the information you could have captured.

In reality for shots that push the limits you end up throwing an exposure based on a middle gray out the window; it's all about exposing to the right, but not blowing a significant amount of highlight ... which will give you the cleanest possible shadow detail when you reveal the shadow detail with ACR.

Keep in mind too that ALL images need post-processing after the HDR compression process.

Colin,

Thanks. I will keep these suggestions in mind the next time I shoot in these backlight conditions. I will also keep the GND filters in mind to help manage the photons.

John

[Colin Southern]

Colin,

Thanks for the suggestion. I have sent you the RAW files now as per your suggestion. Look forward to seeing your next ideas.

Note that I completed the ACR adjustments using the most underexposed image. My intent here was to keep as much color as possible to create a realistic looking sky. Up to this juncture, this has been my most challenging task. I have been able to cope with the noise using tools like Topaz Denoise.

A couple of followup questions:

1) Would you have chosen a different image with longer exposure to complete the ACR?

2) What is your take on management of noise with tools like Topaz Denoise or do you use Photoshop directly for this task?

Thanks again, John

Hi John,

Most of the time I push the exposure until I get just one or two blinkies; as other have pointed out, one can rely on a certain amount of "safety margin" in a RAW capture - but - the sensor response also seems to get a little compressed around this area as well, and I find that if I reply on the recovery slider too much I end up losing sky detail - so personally I push the exposure up to the limit of the histogram on the camera, but not over (keeping in mind that the histogram is generated from the in-camera jpeg conversion and has a bit of an S-curve applied anyway).

With regards to the like of tools like Topaz Denoise - personally - I think they're a waste of money; noise generally isn't an issue with a properly exposed & composed shot - ACR has noise tools built in - and all ANY noise reduction tool ever seems to do is blur detail. THE single best way I know of to address noise is to simply take a number of identical exposures and just average them in Photoshop.

[Colin Southern]

Hi John,

I've processed this from your darkest exposure so as to avoid damaging any more of the sky tones. It looked like the sky was somewhat over-exposed even with the darkest exposure -- probably one of those occasions where a polariser would have helped. I was planning on layering and masking two of the images as I mentioned above, but in the end the fill light control in ACR revealed enough clean detail that I didn't need to. The other thing you might find helpful is www.kelbytraining.com - Matt Kloskowski has an excellent HDR video tutorial (you watch them online), including using Photomatix ... I think if you watch it you'll be able to get much better results with your HDR brackets than you're getting now - they give you the option to sign up for 1 month for $24.95 USD - a pretty good investment (frankly that site is hands-down the best investment ANYONE could ever make to improve their photography & processing, but nobody ever listens to me what I keep saying that, so I don't say it too often anymore! ).

[Jim B.]

Big + for Kelby training.The tutorials are well done and easy to follow.A huge amount of knowledge on that site.