Ad
Ad
Ad
Pages: « 1 2 3 [4] 5 »   Bottom of Page
Print
Author Topic: Synthetic HDR  (Read 25412 times)
Forsh
Newbie
*
Offline Offline

Posts: 27


WWW
« Reply #60 on: April 03, 2007, 03:54:17 PM »
ReplyReply

Quote
Forsh,

Your examples are striking, and I especially like the first image of the young woman. What software do you use to extract the full exposure to the exr and to do the tone mapping?

Bill
[a href=\"index.php?act=findpost&pid=110281\"][{POST_SNAPBACK}][/a]


Thanks Bill, I just batch process Canon RAW to OpenEXR with Photomatix. (default settings) It's really easy to overdo it, so restraint is the key with that program. With this method I get the entire exposure range of the RAW file into one file that can be worked with. After you have the EXR (or HDR) just tonemap it with any program you wish, even Photoshop.
Logged

Futenma Shrine My HDR Photography from Okinawa Japan.  | okinawa japan Other from Okinawa Japan. So what do you do? You don't want create a scene as they can call upon their members beating you down with their home made reflectors in nanoseconds, and creating an international incident over a pix of the rare Zebra butterfly is probably not a great idea.
Ray
Sr. Member
****
Offline Offline

Posts: 8852


« Reply #61 on: April 03, 2007, 08:52:41 PM »
ReplyReply

Quote
Thanks Bill, I just batch process Canon RAW to OpenEXR with Photomatix. (default settings) It's really easy to overdo it, so restraint is the key with that program. With this method I get the entire exposure range of the RAW file into one file that can be worked with. After you have the EXR (or HDR) just tonemap it with any program you wish, even Photoshop.
[a href=\"index.php?act=findpost&pid=110451\"][{POST_SNAPBACK}][/a]

Forsh,
Is there any ultimate dynamic range advantage in this method, or is it just an easier approach? I don't understand why it's not possible to recover the entire exposure range of the RAW file with a single conversion using appropriate adjustments of EC with shadow noise and contrast at zero, plus a bit of curves adjustment and sat. and vibrancy adjustment in CS3 Bridge.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8852


« Reply #62 on: April 04, 2007, 01:12:40 AM »
ReplyReply

By the way, Sarah, that was an excellent April Fool's joke for pixel-peepers   .
Logged
st326
Jr. Member
**
Offline Offline

Posts: 62


« Reply #63 on: April 04, 2007, 01:21:26 AM »
ReplyReply

Quote
By the way, Sarah, that was an excellent April Fool's joke for pixel-peepers   .
[a href=\"index.php?act=findpost&pid=110522\"][{POST_SNAPBACK}][/a]

Eh?
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2756



« Reply #64 on: April 04, 2007, 07:59:14 AM »
ReplyReply

Quote
Forsh,
Is there any ultimate dynamic range advantage in this method, or is it just an easier approach? I don't understand why it's not possible to recover the entire exposure range of the RAW file with a single conversion using appropriate adjustments of EC with shadow noise and contrast at zero, plus a bit of curves adjustment and sat. and vibrancy adjustment in CS3 Bridge.
[{POST_SNAPBACK}][/a]

Forsh and Ray,

One obvious difference in Forsh's method is that he is converting from raw using the Photomatix software and then editing in a 32 bit floating point linear space. If you are using Bridge and ACR, the internal working space is 16 bit linear. On an intuitive level, it would seem to me that the ACR 16 bit linear space would have all the headroom necessary to process the 12 bit linear raw file. If you have a true HDR image, then 32 bit floating point would be necessary.

I note that in PSCS3 one can use the mode command to convert to 32 bit and save the file in the OpenEXR format. According to a [a href=\"http://www.anyhere.com/gward/hdrenc/hdr_encodings.html]post[/url] by Greg Ward, the OpenEXR format file can use single precision (32 bit) IEEE floating point or a half precision (16 bit) format for storage, but 32 bit IEEE floats are presumably used for processing, since that is the format supported by current CPU hardware. Some graphics cards now support the half precision format in hardware. PSCS2 can't store in the OpenEXR format, but it does support 32 bit floating point and I think that the PSD or Radiance formats that it does support for storage would be equally suitable.

It would appear that PS users could use Forsh's methods without investing in the Photomatix software, but I have no idea of which workflow would be better. Or perhaps one could get similar results from ACR4 as Ray suggests. An interesting topic for discussion.

Bill
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8852


« Reply #65 on: April 04, 2007, 08:30:20 AM »
ReplyReply

Quote
Eh?
[a href=\"index.php?act=findpost&pid=110525\"][{POST_SNAPBACK}][/a]

You mean it wasn't an April Fool joke? You did post the link to the plug-in on April 1st. Was that just a coincidence?  
Logged
rslv
Newbie
*
Offline Offline

Posts: 4


« Reply #66 on: April 04, 2007, 09:58:42 AM »
ReplyReply

Sarah, thank you for creating and sharing this. Much appreciated.
Would you mind posting a workflow example? I'm not entirely sure about your process.
Logged
Pete JF
Full Member
***
Offline Offline

Posts: 185


« Reply #67 on: April 04, 2007, 10:32:53 AM »
ReplyReply

Man, I'm interested in some light HDR technique but this thread has left me spinning. My terminology/knowledge is lame in this area and I do not understand what is going on here. What Forsh is doing looks pretty good. My needs are simple..I'd just like to open up some areas in certain mages in a simple subtle way. I'm interested in what, you, Sarah, are presenting here but I'm not sure I understand what is different.

Obviously Forsh is using one image, one exposure..I don't understand what happens after that. Are you creating several interpretations of the exposure in a RAW processor and then outputting those for HDR treatment... Anyone care to explain in less terminology oriented terms?




Some of the HDR stuff I've seen draws WAY to much attention to the processs.
Logged
Eric Myrvaagnes
Sr. Member
****
Offline Offline

Posts: 7789



WWW
« Reply #68 on: April 04, 2007, 07:19:51 PM »
ReplyReply

Quote
Man, I'm interested in some light HDR technique but this thread has left me spinning. My terminology/knowledge is lame in this area and I do not understand what is going on here. What Forsh is doing looks pretty good. My needs are simple..I'd just like to open up some areas in certain mages in a simple subtle way. I'm interested in what, you, Sarah, are presenting here but I'm not sure I understand what is different.

Obviously Forsh is using one image, one exposure..I don't understand what happens after that. Are you creating several interpretations of the exposure in a RAW processor and then outputting those for HDR treatment... Anyone care to explain in less terminology oriented terms?
Some of the HDR stuff I've seen draws WAY to much attention to the processs.
[a href=\"index.php?act=findpost&pid=110592\"][{POST_SNAPBACK}][/a]
Me too! I keep wondering whether you can get the same results just using the highlight-shadow tool.
Logged

-Eric Myrvaagnes

http://myrvaagnes.com  Visit my website. New images each season.
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #69 on: April 04, 2007, 08:07:18 PM »
ReplyReply

Try this simple technique.

Process the original RAW 2 times, adjusting the exposure - one over exposed to bring out the shadows and one underexposed to reduce highlight clipping.

Open the under exposed image and copy the over exposed image on top.

With the top layer selected (the over exposed one), hit ctrl alt ~  this selects the brightest pixels of the over exposed layer, then Delete.  (after ctrl alt ~ you could hit the mask icon and go from there).

The problem doing this in PS HDR with one image is that you have to edit the exif to fool PS into thinking there are 2 seperate exposures.

In general, using lightroom, I don't have to do this, but the original sample from Sarah was a bit extreme and so this was easier than fussing with the curves and highlight/shadow stuff in LR.
Logged
pobrien3
Sr. Member
****
Offline Offline

Posts: 320


« Reply #70 on: April 04, 2007, 09:47:55 PM »
ReplyReply

By processing the RAW several times and blending them as if you had a true HDR image, you can achieve the look that Forsh gets which is decidedly different than using one 'development' of the file and using curves / shadow/highlight.  I don't say better or worse, just different.  I personally like to use this method for some images to boost local contrast before converting to B&W - you can get quite a dramatic effect: I like what it does to skies and clouds, and you can use it to boost textures.  If it's subtlety you're after, this technique must be used very sparingly and with great care - in such cases I would try PS curves etc. first.  Below are a couple of very overdone images to illustrate the effect.

There's an excellent tutorial on using this technique (and some stunning photos) with Photomatix by Pete Carr here - I particularly like the images of my home city, Liverpool on his site.

[attachment=2236:attachment]  [attachment=2237:attachment]  [attachment=2238:attachment]
« Last Edit: April 04, 2007, 10:09:41 PM by pobrien3 » Logged
neil
Administrator
Sr. Member
*****
Offline Offline

Posts: 8923



WWW
« Reply #71 on: April 04, 2007, 10:48:03 PM »
ReplyReply

I'd be interested in your porting your plugin to a mac....

I'm not sure if I have the time for 32bit floating point something or other....but I'll look into it - cause I want to get a similar, but smoother look without having to shoot polaroid spectra.

I developed a faux HDR workflow using Dr Brown's Place a Matic and the attached actionset. [attachment=2240:attachment]   Its pretty handy and customizable at each stage of the workflow, cause sometimes a linear selection of the highlight or shadow conversion isn't right.  So you can alter the development settings in ACR while altering mask tonalities and the mix between the two files with custom dodging and burning. The actionset finishes with some pretty handy tricks to get midtone contrast returned.  I'd love to hear if you experts have any suggestions for improvement iworkflow steps in the actions to give better tonal qualities.

I developed the need for a quick and accessable workflow after shooting in some hard desert light:


« Last Edit: March 20, 2008, 07:01:41 PM by neil » Logged

st326
Jr. Member
**
Offline Offline

Posts: 62


« Reply #72 on: April 04, 2007, 10:55:32 PM »
ReplyReply

OK, the basic workflow (and no, it wasn't an April fool's joke) is something like:

1. Capture image, get it into Photoshop by some appropriate means.

2. Make sure that the image is in 16 bit mode.

3. Run the plugin. If you can see any difference to the image as-is, assuming a typical 8-bit per colour plane graphics card, you've probably got the plugin set to be too aggressive. You should only see any difference if you up the contrast significantly. I keep reminding people that the point of the plugin *isn't* noise reduction, though it does do that to some extent. A good trick is to zoom in on some shadow detail, create a (temporary) levels layer and set it so that the shadow area is very visible. Tweak the plugin to get smooth shadow detail, but no effect to the overall tonality or sharpness of the image.

At this point, it kind-of depends on what you want to do. If you want to do tone mapping, put the image in 32 bit mode then back to 16 bit mode again, which will cause the tone mapping dialogue to appear -- you'll need to select Local Adaptation from the drop down box. This will let you perform the kind of tone mapping that's popular in the HDR community -- it's annoying that you can't just invoke it directly on a 16 bit image, but it just seems to be the way PS is designed.

One thing that's interesting with this plugin is that it seems to behave quite differently depending on what kind of source material you feed it, so I think I'll have to add a few more tweakable controls to the next version. I've tested it mostly with images from my Megavision monochrome back and from my Better Light scan back -- neither of these things have a Bayer matrix, so there is none of the usual colour interpolation going on down at the pixel level. I tweaked the code based on my own images, and set it up so that when it found fine detail it backs off, leaving it unchanged. However, it seems that the difference from pixel to adjacent pixel in both the Better Light and the Megavision images is substantially larger than that from interpolated colour data from Bayer matrix sensors, so in this case there wasn't anything like as much fine detail as I was seeing from my test images. I suppose this shouldn't be any surprise, because this is exactly what theory would predict, but whatever the cause I'll have to add some kind of fiddle factor for this tendency. This difference isn't so visible to the naked eye, but it's certainly there in the numbers, and the plugin works on the numbers. Strictly speaking, the red and blue channels have half the resolution in each axis (i.e. only 25% of the information), and the green channel has only 50% of the information of a non-Bayer image, even ignoring the effect of antialiasing filters, so this significantly limits the maximum slew rate between adjacent pixels, even when that slew is just caused by noise.

I suspect that two quite different algorithms are going to be necessary -- one for non-Bayer imagers, and one for conventional imagers.
« Last Edit: April 04, 2007, 10:57:54 PM by st326 » Logged
st326
Jr. Member
**
Offline Offline

Posts: 62


« Reply #73 on: April 04, 2007, 11:09:01 PM »
ReplyReply

On the subject(s) of tone mapping and noise reduction, I'm going to have a go at putting a plugin together that does image manipulation in the spectral domain (i.e. take the source image, do a 2D FFT on it, mess with the spectrum, then do an inverse FFT to map it back to a conventional image). Lots of standard operations on images become trivial in the spectral domain -- sharpening, blurring, etc. are trivial, because they map on to something not much harder than cropping or at worst dodging and burning in the spectral domain. However, there's a lot more that can potentially be done -- transforming the spectrum nonlinearly should make it possible to do spectral noise reduction (as commonly used in the audio world) on a 2D image. I've no idea how well this will work in practice, but there is a good chance that it would be extremely effective. I've written audio noise reduction code (in a past life) that could pull intelligible speech out of something that sounds like 99% hiss, 1% signal, so given the amount of noise we usually have in images it's probably going to work pretty well, if it works at all.

Interestingly, tone mapping has a parallel in the spectral domain too. In effect, with tone mapping, you're reducing the contrast of large scale features in the image, whilst simultaneously increasing the contrast of smaller scale features. In the spectral domain, this is no harder than dodging and burning (or maybe applying curves). Should be a fun plugin anyway. It might also provide an alternative means of upping effective bit depth for Bayer images, because that too should be a lot more precise in the spectral domain.

(By 'spectrum' and 'spectral' I'm referring to the signal processing terms -- this is completely unrelated to the colour spectrum of the image itself)
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8852


« Reply #74 on: April 04, 2007, 11:15:28 PM »
ReplyReply

Quote
Try this simple technique.

Process the original RAW 2 times, adjusting the exposure - one over exposed to bring out the shadows and one underexposed to reduce highlight clipping.

Open the under exposed image and copy the over exposed image on top.

With the top layer selected (the over exposed one), hit ctrl alt ~ this selects the brightest pixels of the over exposed layer, then Delete. (after ctrl alt ~ you could hit the mask icon and go from there).

The problem doing this in PS HDR with one image is that you have to edit the exif to fool PS into thinking there are 2 seperate exposures.

In general, using lightroom, I don't have to do this, but the original sample from Sarah was a bit extreme and so this was easier than fussing with the curves and highlight/shadow stuff in LR.
[a href=\"index.php?act=findpost&pid=110697\"][{POST_SNAPBACK}][/a]

Tim,
You're a gem. I just tried your method on a couple of 'real' HDR exposures, the view out of my $4 a night hotel room in Nepal, bracketed, and the halos around the edges of the window frame have disappeared. Michael should include this trick in his tutorial on blending.

[attachment=2241:attachment]


I've just realised I forgot the Gaussian blur in the above blended shot. Must have had a few too many glasses of wine when I posted that. I have great difficulty in avoiding halos at the edges of the window frames in shots like this. Setting a high pixel radius in Gaussian Blur helps but creates other 'lack of contrast' and 'uneven contrast' problems in the view out of the window. I was hoping Tim's method was a quick fix but it doesn't seem to adress this problem.
« Last Edit: April 05, 2007, 08:01:35 PM by Ray » Logged
Slough
Guest
« Reply #75 on: April 05, 2007, 07:19:02 AM »
ReplyReply

There is another way to 'recover' highlights and shadow detail using only one image. Obviously you cannot recover anything that is not there, so really this is contrast masking. Here is the tutorial:

http://www.nwpphotoforum.com/ubbthreads/in...ston/roman2.php

I find this can work really well, and it does not unduly flatten the image i.e. it preserves local contrast.

I would prefer an automated method as the above one requires some thinking, and hence takes some time, but it does the job.
Logged
st326
Jr. Member
**
Offline Offline

Posts: 62


« Reply #76 on: April 05, 2007, 12:35:53 PM »
ReplyReply



Split Rock, Joshua Tree National Park

Cambo Legend 4x5, Better Light Super-6K, 8000x6000x14-bit colour, single capture (approx. 300MB). The original has extreme contrast -- the sun was directly behind the rock, just before sunset. Processed as follows:

   1. Synthetic HDR plugin, 3% amount, strong noise reduction.
   2. Changed to 32-bit mode.
   3. Changed back to 16-bit mode with local adaptation, 1 pixel radius, gamma at maximum, curves unaltered from default (i.e. linear)
   4. Created levels and curves layers, tweaked until it looked somewhere close
   5. Created a colour balance layer, tweaked to remove a slight blue cast
   6. Created a second curves layer, tweaked to darken the sky slightly. Hand-painted mask, then blurred mask with a 250 pixel Gaussian blur filter.
   7. Did mild sharpening with Smart Sharpen (it didn't need much)
   8. Duplicated the image layer, applied a 150 pixel Gaussian blur, then set the layer to affect colour only. Created a mask, inverted it, then painted over areas of the plants that had motion artifacts due to fairly strong wind.
   9. Crop, copy, downsize for web use, slight unsharp mask to help compensate for the (huge) size reduction.

What looks like it might be noise in the rock surface areas isn't -- it's actually the grain of the rock itself. Downsizing the image for posting here meant reducing both dimensions by roughly 10:1, which is a 100:1 reduction in information, so at this size, if you can see it, it was there.
Logged
pobrien3
Sr. Member
****
Offline Offline

Posts: 320


« Reply #77 on: April 05, 2007, 12:40:36 PM »
ReplyReply

Slough, if you go for this method there are better ways of making the mask than using the colour range selection tool.  The technique Tim Gray described above is one such method, using luminosity masks is another - the one I prefer.  Take a look at this tutorial by Tony Kuyper, who also kindly provides the actions required to make the masks.

Peter
Logged
Slough
Guest
« Reply #78 on: April 05, 2007, 01:49:12 PM »
ReplyReply

Quote
Slough, if you go for this method there are better ways of making the mask than using the colour range selection tool.  The technique Tim Gray described above is one such method, using luminosity masks is another - the one I prefer.  Take a look at this tutorial by Tony Kuyper, who also kindly provides the actions required to make the masks.

Peter
[a href=\"index.php?act=findpost&pid=110828\"][{POST_SNAPBACK}][/a]

Thanks. Roman's works well but I will give your link a try. Leif
Logged
rslv
Newbie
*
Offline Offline

Posts: 4


« Reply #79 on: April 05, 2007, 05:23:06 PM »
ReplyReply

Sarah, thanks for posting the workflow details.
Your image rocks!  
Logged
Pages: « 1 2 3 [4] 5 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad