I think I was pretty clear that I was not talking about dedicated film scanners. The Minolta is one of the few scanners I have not tried. What I am talking about is a mechanical issue on the flatbeds as it varies depending on the quality and age/usage of the flatbed scanner and varies tremendously depending on whether both scans are done on a warm scanner. One of the ways to defeat the problem is to do a batch scan with a couple of dummy scans up front to warm up the scanners. The latter batch scans register much better. It has become part of my workflow, when I need a high res scan, to do a warm up dummy scan first before the real multi-exposure scan on an Epson 750 or a Microtek 1800f. I have done some limited beta testing with Sf on this issue. If the scans are too far out of wack the software cannot bring it back in, but if they are a little it can handle it-hence the warm up scan. The latest version of SF has extented the "out-of-wack" parameters it can pull back in, but one must work very carefully to take advantage of it. Doing wet scans also helps with registration as there is no issue with the film changing shape from temperature differential. It is similar to having a negative "pop" from the heat as in enlarging resulting in the center of the scan being off register.

I'm thinking this should be pretty easy to solve on a cheap scanner too. Instead of using the instable step motor to do two scans one could just expose each step twice, then do a step, expose twice, do step etc...
Then the two scans would be interleaved in the same picture which is very easy to separate..

:-) I just wrote the same and now got to read your post..

The bulk of this thread is about combining multiple exposures,either using PS or using the cool routine developed by GLuijk. My thought is; put the routine in the camera and let its do it all.

I was thinking along the lines of the camera taking bracketed (ISO) exposures and doing the pixel by pixel selection and replacement using the technique decribed at the start of this thread. Only the final combined image would be saved to the memory card. Perhaps the user would set two ISO's for the exposures. One ISO gets the highlights, the other the shadows. The shutter speed and f stop would have to be the same for each exposure to avoid having wierd things happen to the combined image.

It wouldn't surprise me at all to find out this could be done with nothing more than a firmware update to the camera.

Fuji S3 is doing something similar, having 2 interleaved CCD arrays of different sensitivity.

I think there will be more "weird things" happening to the combined image from ISO bracketing than shutter speed bracketing.

But yes, this could be an attractive feature to some photographers, though probably not to those who like to shoot in raw.

You are really reinventing the wheel here. Blending together the best parts of frames shot with different exposure levels has been around several years prior to HDR blending being added to Photoshop as a feature. I've been doing so since 2001 or so when I got my first digital camera.

Your technique can be easily duplicated in Photoshop without any special software plugins. Shoot a frame with normal ETTR technique (don't blow the highlights), and a second frame with +3 stops exposure. Process both RAWs with identical settings except for exposure, the second one processed with exposure set to 3 stops below the first. Stack the second exposure on top of the first in Photoshop, with layer blending set as follows:

Click to view attachment

Split the two halves of the "This Layer" white slider apart by holding down the ALT key, position them as shown, and you're good to go. This will make luminance values of 220 or above come from the normal exposure, and luminance values of 35 or less to come from the +3 exposure layer. All intermediate luminance values are blended from both layers so there is a smooth transition. Setting the white blend point to 220 keeps saturated colors with clipped channels from contaminating the highlights, and seting the black blend point to 35 keeps the worst of the shadow noise from having any effect at all on the final image. Having a 185-point luminance blend range makes sure that there are no visible seams in the blended image. You can save the layer blend settings as a preset for convenient later use.

I've used this technique successfully for many years.

Unfortunately, it doesn't work well at all with the sample raw files provided by Guillermo Lujik. There are nasty highlight artifacts. Granted, the difference is 4 stops, not 3. Adjusting the sliders so that one gets rid of the highlight artifacts also seems to get rid of the reduced noise.

Care to give it a try, and compare your results with his?

Here's the post with the link to the raw files.

The post has a link to his article, which has a link to a forum discussion with no RAW links. Post a link direct to the RAWs, so I can find them.

If you're talking about the processed JPEGS, you can't use those because they havent been equalized to matching tonal levels. For my technique to work, you must process one RAW at exposure 0 and the other at exposure -4 (if the shots were bracketed 4 stops apart), otherwise the blended result will look very weird.

The actual exposure settings don't matter, as long as they are the correct interval apart. Given a 4-stop bracket, if the normal exposure is best processed at +0.5EV in ACR, then the other RAW should be processed at -3.5EV. But if the normal exposure is best processed at +0.1EV, then the other RAW should be processed with a -3.9EV exposure setting.

They're one the first page of this thread. I'm very interested to see if the same results can be achieved in Photoshop as well. I'm going to try it right now....and it does seem to do a very good job indeed!

jdv

Click to view attachmentClick to view attachmentClick to view attachment

Jani,
Jonathan's method seems to be an improvement over other blending techniques I've come across, but it still relies upon post-blending skills in Photoshop.

I struggle with Photoshop techniques. I know I should spend hours and hours pouring over Photoshop manuals, but for some reason I'm reluctant to do so. Is it a fear of becoming a nerd, or is it just I'm a lazy student who wants the glamour but not the hard yakka? I don't know. A bit of both probably.

I mentioned before, in blending images there's often a halo effect around borders between high contrast transitions. I don't have the skills to get rid of these, at least without painstaking hours of work.

I'm thinking that maybe GLuijk's method offers a quick solution to this.

If you've got the patience, I'll go through a processing procedure I used to blend a couple of images that varied by 4 stops, using Jonathan's method. The final result is not perfect, and not to my complete satisfaction.

The following is a scene of a 5 star hotel 'ensuite' in the foothills of Nepal. (5 stars is relative, you understand ). The cost for one night in this luxurious accommodation was just US$4. That's less than 1$ per star . I wanted to capture the scene out of the bathroom window, as well as the bathroom itself.

One shot would simply not do. So I brought my tripod into the bathroom and took a series of 3 bracketed shots.

Here's my attempt at blending 2 of those 3 shots, using Jonathan's method.

(1) The initial blend. Jonathan's recommended settings for "This Layer" should be varied according to the circumstances. It might be right for exposures differences of 3 stops, but my exposures differed by 4 stops, consequently my split layer values were not 35/220, as Jonathan suggested. but 28/43, based upon eyeballing, not mathematical calculation.

Click to view attachment

Having flattened the image, I then proceeded to use the limited but familiar techniques I normally use to get the image into shape.

First, a curves application to raise the shadows.

Click to view attachment

Then a shadows/highlight adjustment.

Click to view attachment

Then a levels adjustment.

Click to view attachment

After the levels adjustment (ctrl click on RGB channels, reverse selection, layers levels, 80% opacity), the levels histogram looks like this.

Click to view attachment

So here is the rather flat image lacking in umph!

Click to view attachment

The problem is; how do I get from this rather flat image to the final vision without introducing artifacts around the edges of the window frames?

Click to view attachment

I think what a lot of you are missing here is that this technique is not making an image where you actually see the HDR range increased, but reducing noise in the shadow areas, and bringing back detail, like nothing I've seen, like RAW's "lluminance" etc smoothing, or even PS's "Reduce Noise" filter. I may be wrong, but what hit me is the incredibly clean shadows in what appears to be a a typical dynamic ranged image.

As for automation, just wait. I'm quite sure this will become standard in cameras simply because getting rid of noise is like DUH! No noise is good noise, and if you want noise, then simply switch off advanced noise reduction in camera.

Does this work for you, Ray?
Click to view attachment

I'm not trying to knock what is being discussed but it really isn't new and has been discussed several times e.g.
Noise Reduction with multiple exposures

If you seach through the forums there are examples of using HDR to reduce noise as well as image stacking and we have discussed a number of software packages to reduce noise and increase dynamic range on a number of ocassions.



If you have halos around high contrast edges that you need to get rid of then I would suggest shooting multiple exposures no more than 2/3 or 1-stop apart. Blending in Photoshop (and I guess any other technique) will mitigate the effect of halos provided you have sufficient exposures from which to extract 'clean' data. Two exposures 4-stops apart doesn't give sufficient information to eliminate all types of artifacts.

Yes, this appears to accomplish with a bunch of exposures what this technique manages with two, and at the cost of more than a little bit of extra manual labour. The exception may be the benefits of the image stack program, which appears to introduce additional flexibility.

For those who don't want to follow links twice, here's a direct link to the PDF (1.7 MB).


From what I've seen and recall, none appear to have the same simplicity as this method.


Why shoot four exposures 2/3 or 1 stop apart, when you can settle for two exposures three or four stops apart?


It seems to work well enough in the examples provided in this thread; I see no artifacts in the images presented to us.

Perhaps more rigorous testing will reveal such artifacts, but so far, Guillermo's method seems better for the sheer simplicity of the procedure.

He's followed the KISS principle.

Jonathan,
Not quite, but thanks for the attempt. With that amount of light streaming through the window, the eye would pick up a much brighter scene around the vanity basin. Below is what I'm aiming for. But to achieve this, I had to carefully isolate the scene out of the window with the polygonal lasso tool and process the two parts of the image as though they were separate images. The foliage is perhaps a bit too saturated, but this is the sort of effect I want.

However, in a situation like this, if I have to use the lasso tool, I could simply copy & paste the 'window view' from the dark image to the light image. With both methods I have the problem of that transition edge along the window frame, a problem which is not particularly apparent in the jpeg but is definitely there as can be seen in the crop.

Click to view attachment Click to view attachment

David,
This is perhaps the appeal of GLuijk's method, when fully developed. Two exposures 4 stops apart should contain all the data one needs.

Here's a way to fix your edge blending problem:

1. Have the window view and interior in separate layers, with window view layer on top.

2. Create a layer mask for the window view layer.

3. Assign the original window view exposure (where the room view is very dark) to the layer mask.

4. Do a level adjustment on the layer mask to make it straight black and white with no intermediate values, so that the window view is 100% white and the room view is 100% black. Depending on image content, you may need to do a bit of paintbrushing to eliminate stray specks where they don't belong. The Dust & Scratches filter can be handy for this, but must be used very carefully.

5. Do a 2-pixel Gaussian blur on the layer mask to make the transition soft and fuzzy. The edge will look horrible, but don't panic, there's a method to the madness.

6. Do another Levels adjustment on the layer mask. By adjusting the source white and black point sliders, you will be able to move the outer and inner edges of the blend transition zone independently anywhere within the 2-pixel blur zone, which will avoid matte lines and make the blend as seamless and natural-looking as possible.

Whenever you're compositing images, you always should have a bit of fuzziness at the edge where one layer transitions to another, or you'll either have matte lines or an artificial "cut-out" look to the edge. This technique makes it easy to get natural-looking blends along composited edges.

Yes, I realise this. It's really an issue of how much stuffing around one needs to do to achieve the right balance which looks natural. However, I'll try to go through those procedures you've outlined tomorrow with a clear head.

I noticed that the CS3 demo version had a feature whereby one can enlarge or diminish a selection by a specified number of pixels, as well as specifying a degree of feathering. That could be useful with this particular image.

Out of curiosity, I loaded the 3 RAW bracketed images of this scene into CS2's HDR. The results are at least equally useful. There are many ways to get a desired result. I used a bit of dodge and burn along the window edges. It's probably not an ideal image to demonstrate the halo problem because I see signs of clear silicon sealant that have smeared the edges of the window panes .

Click to view attachment

The advantage to my technique is that it allows one to expand or contract the selection, as well as adjust feathering amount of the blend, with a real-time preview.

I noticed the sealant on the window as well. The small upper section of window visible seems to be free of it though.

On a different note, I spent a few hours playing around with some bracketed triples 3 stops apart, experimenting with various blending techniques. I discovered that using all 3 images works best. Blending the +3 and -3 shots only leaves somewhat of a quality gap where the highlights and shadows look good, but the midtones are noisy, and the transition between the highlight shot and the shadow shot is uglier. The comparison crops have all had the same sharpening:

2 frame blend: (+3 and -3 only)
Click to view attachment

3 frame blend: (+3, 0, and -3)
Click to view attachment

As you can see, the highlights and shadows look equally good (as they should, coming from the same frames), but the midtones of the 2-frame blend are much noisier.

The attached ZIP file has the layer styles I used to blend the frames. For the 2-frame blend, the -3 frame was on the bottom as background, and the +3 frame was on top with the "HDR Shadow Layer Blend" style applied. For the 3-frame blend, the 0 frame was background, then the -3 frame with the "HDR Highlight Layer Blend (3 Frame)" style applied, then the +3 frame with the "HDR Shadow Layer Blend (3 Frame)" style applied. After flattening both, I picked a crop area that had samples of the entire tonal range, cropped, and applied equal sharpening to both.

To use the blend styles, simply unzip and then load the extracted file from the Styles palette in PS. Process all 3 RAWs exactly the same except for the exposure setting. The 0 file is your best guide for things like setting white balance, lens corrections, etc.; try to adjust exposure of the 0 file so that shadows and highlights are about equally clipped. If you only want to improve tonality within a single-frame exposure range, be strict about setting the exposure of the -3 file exactly 3 stops above the 0 file setting, and the +3 file exactly 3 stops below. But if you want to include more DR in the frame (which will of course make it look flatter after blending and require more curves and local contrast boosting), adjust the -3 file for the best highlight appearance (no or only light specular clipping) and the +3 file for best shadow appearance (no or only light specular shadow clipping). Then stack the layers 0 (background), +3 / HDR Highlight Layer Blend (3 Frame), -3 HDR Shadow Layer Blend (3 Frame).

Enjoy!

The ZIP file downloads as "index.php". Did get it to work by renaming to "index.zip". Then it would unzip to HDR Blend States ASL.

Thanks for the great hints......

Jim

Yeahh! we did it! a mate of me and me have modified DCRAW's source code to implement this technique as an extra option. And it works great!!! it has some advantages over the little test program I provided here, as it works straight from the RAW files. I mean it does the image blending before the RAW developing stage and that has many important advantages:

1. Noise free virtual RAW: Bayer interpolation is done over a 16-bit "virtual" RAW where noise has already been minimised. This way we avoid noise propagation into interpolated cells.
2. Process previous to white balance: WB rescales all channels, so the threshold value for the blending algorithm had to be chosen for the most severely corrected channel. Now an optimal threshold is chosen for the three channels at a time as WB has not been applied yet.
3. Only one RAW development: thanks to the internal virtual RAW data generated, no matter how many files are to be blended, only one RAW developing will take place incresing calculation speed.

For those familiar to DCRAW the miX option will be something like:
dcraw -X [threshold] [exposure correction] ... *.cr2 for instance
It will take all indicated RAW files, calculate their relative exposition no matter which it is, and blend them into an optimum noise free image with the same appearance (contrast and tone) as the original but with expanded dynamic range in the shadows and zero noise.
As people are not used to DCRAW perhaps I will write a simple GUI for this tool.


Hi Jonathan Wienke, just to say that I never stated that reducing noise through multiexposure was a new idea; I am sorry if it looked like that. In particular in my website I textually say: "Las ideas descritas hasta ahora no son nuevas, la novedad consiste en aplicarlas con el fin de obtener una reducción de ruido radical y de forma automatizada (...)", that means "The ideas described here [referring to the noise reduction process through overexposition] are not new, the new thing consists of applying them with the goal of achieving a radical noise reduction in an automated way".

If you can get the same noise reduction as this modifed DCRAW with PS, at the same speed (about 10s to process 3 shots starting from their RAW files), and with the same ease of use (one button click), just let me know. Otherwise you are invited to try our tool.
Apart from the advantages of RAW blending commented at the begining, a progressive blending technique as the one you propose in PS is not optimal (nor needed) compared to a more simple but effective pixel selection algorithm for two reasons: first your noise reduction is not maximised as you are mixing noisy pixels with noise-free pixels, while I simply discard the first ones. Secondly, even minimal image overlap errors (less than 1 pixel) would lead to a loss of sharpness due to averaging in your case.


I did a test yesterday over a high contrast scene (about 13 f-stops of real dynamic range) and achieved a completely free of noise histogram with texture recovery in the darkest shadows:
(please note these histograms are logaritmic plots where each vertical division corresponds to 2 image f-stops. The original hardly achieved to represent properly 8 f-stops due mainly to noise and quantization. The blended version clearly increases the histogram precision up to 13 f-stops):




(what can be seen on the floor is not noise but the tiles texture)


Please find here the RAW files of my two test images for anyone wanting to try alternative techniques:

Couch:
http://www.guillermoluijk.com/download/sub.cr2
http://www.guillermoluijk.com/download/sobre.cr2

Blue room:
http://www.guillermoluijk.com/download/foto1.cr2
http://www.guillermoluijk.com/download/foto2.cr2
http://www.guillermoluijk.com/download/foto3.cr2

That's fantastic!

I'm not familiar with DCRAW but don't mind working on the command line. Is there somewhere I can download your version? Will this be able to output a DNG file?

it's still not ready, coming soon.

we were thinking about that, to produce a 16-bit DNG as an output. Would be orgasmic (lol). But need to know a lot about RAW file formats. Perhaps with assistance from David Coffin...

Well DNG would be a quite welcomed feature!

I second that!

My Spanish is very limited, so I wasn't able to read your web site article. My apologies. The first program you posted about breaks no new ground, and offers little or no advantage over blending techniques that have been available for years, including the method I posted about using layer blend styles.

On the other hand, your DCRaw modification is very interesting and is something new, DR blending the linear RAW data prior to other processing. This approach is much better than blending after gamma adjustment and tone curves have been applied.

If you can create a tool with a GUI that can take 3 RAWs exposed at 3-stop intervals, blend them, and output a blended 16-bit DNG that ACR or any other DNG-aware program can process the same as the original RAW, I'd happily buy it. It would need to meet the following criteria:

1. The highlights would need to be set so that the brightest value in the darkest image is scaled automatically to the maximum usable value in the output file without clipping.

2. The output DNG must process identically to the original RAW. This means that the normal ACR color processing must work identically to an original RAW. I've color calibrated ACR to my cameras, and would not want to get different color when processing a blended DNG than when processing a single RAW. ACR would have to recognize the camera model and serial number from the DNG file for this to work. The complete DNG file format specification is available from Adobe, so this shouldn't be too difficult. The hardest part would be figuring out what shutter speed to put in the metadata when the file is derived from several exposures with different shutter speeds.

3. Some means to manually adjust registration between the RAWs would be necessary, expecially if it could be done in 1/2-pixel increments. This would require that Bayer interpolation be done prior to blending. The output file would have to be a linear-RGB DNG, but ACR's normal color processing would not be affected. Limiting output to 16-bit linear RGB would greatly simplify your work; you'd only have one output format to worry about. It is just tagged with the appropriate camera model, serial number, and other metadata. DCRaw can handle the plethora of input formats, so you wouldn't have to re-invent the wheel for that side.

If the Bayer interpolation routine was modified so that it took all source images into account simultaneously, slight mis-registration would actually be an advantage, because there would be more than one color channel at each output pixel site. Imagine that when shooting, the +3 exposure was shifted 1 pixel vertically relative to the 0 exposure, and the -3 exposure was shifted 1 pixel horizontally relative to the 0 exposure. After registering the RAW data, the Bayer interpolation now has 2 color channels per pixel to use, either red and green, or green and blue. The exposure scaling would have to be accounted for, but increased color accuracy could be achieved. Some medium format digital backs do this (but not at different exposure levels) to improve color accuracy; 3 exposures are taken, whith the sensor being moved 1 pixel vertically or horizontally between frames.

You can download the complete DNG specification from a link at the bottom of this page.

Edit: The Bayer interpolation modification wouldn't really be necessary, the color accuracy improvement happens already when the different exposures are blended together. Never mind.

The program I already posted blends the images prior to any gamma correction, as DCRAW is a linear RAW converter (if used in -4 mode, and I did). The improvement of this new DCRAW version is that the blending is applied even before the white balance, which improves even more the accuracy of the threshold set and in addition to this saves some RAW developings. We can say it's more elegant.
This is a comparision between the results of:
1. My previous program and a conservative threshold decision value: still noise on some areas (wall).
2. My previous program and an agressive threshold decision value: noise is almost gone but incorrect colour pixels appear (couch).
3. Modified DCRAW with a conservative threshold decision value: noise is almost gone and more regularly spreaded, and no colour artifacts.




We don't sell software. All we do if for free and for fun. It's not likely that we will achieve the DNG output as it's an effort beyond our intentions, and we love DCRAW! so we trust it. DCRAW is a really VERY good RAW developer itself. But I will check your links to find out, thank you.


"1. The highlights would need to be set so that the brightest value in the darkest image is scaled automatically to the maximum usable value in the output file without clipping."

This requirement would be fine in theory. In practice it is not so much, and the reason is that the least exposed imaged provided to the program is taken as a reference for the output. The improvement of scaling it so it fits the maximum (65535 level) is not worth as almost ALL valid images have some pixels close to blow or directly burnt (this is the usual case). You can use this program of mine: Histogrammar that plots 16-bit histograms to check this by yourself on any of your images. If you wait a couple of days, I have developed a new version of Histogrammar which reads directly image files (the one currently in the link needs some mess with a PS plugin and translates the image into a text file. horrible but I didn't have the libraries needed to read 16-bit images) and even plots logaritmic histograms like the one I showed before.
However we will introduce an exposure correction parameter into this DCRAW, so that: dcraw -X 4.0 ... *.cr2 for instance will correct by 2 f-stops up (2^2=4.0) the resulting image avoiding level agregation and mathematical rounding errors if that would be done afterwards.

Moreover the white balance can be very agressive. A scaling of more than one complete f-stop in some channel is very usual and this burns pixels; but few of them. So it is strange to find an image that can strictly be exposed up as most of them have something burnt (and if they don't have, is because were heavily underexposed).
The important thing is to have a good exposure gap between the least and the most exposed images. This really fulls with detail all the shadows thanks to the 12-bit to 16-bit conversion. These images can handle really severe toning curves before showing any banding or posterization.
Our program calculates itself the relative exposures between images without looking at the metadata; and handles N images with N from 2 to what your memory can store, taking always for each individual pixel the best possible value from the whole set of images provided. In fact, this relative exposure calculation is more accurate than using the EV values set in the camera since lighting conditions could vary slightly during the shots (for instance a cloud passing by).

Thanks for taking your time, we will let you know when this DCRAW is ready.

I've been working with sets of 3 images exposed 3 stops apart, +3, 0 and -3. The -3 image usually doesn't have any clipped pixels unless there is a light source in the image, like a bare light bulb or something really shiny with bright specular highlights. Blending these images together with exposure set to -2, 0, and +2 gives a slightly flat-looking initial result, but which still responds well to curves and local contrast enhancement so that you get nice smooth transitions to clipped highlights and shadows, and end up with about 2 stops more DR in the image as well as practically no noise throughout the tonal range.

So I'd like to be able to boost that -3 image to max values below clipping, and scale everything else below that. I'm not talking about changing the WB settings in the RAW, scale all channels by the same amount so the most exposed channel maxes out at 65535 and everything else scales proportionately. When you add 6 stops to the 7 stops or so that most cameras can capture well in a single exposure, using all of the 16-bit linear scale is not a bad idea, especially given that the blue channel can be 2 stops or more below the red channel in incandescent and similar lighting. That means the deepest shadows in the blue channel can be 15 stops below the highlights in the red channel, and an unused stop of highlight room can mean the difference between whether that shadow detail is processed meaningfully or gets posterized into oblivion.

oh I understand now. Your way to proceed is to bracket a couple of stops below and above the "correct" exposure. In my opinion bracketing underexposed as a rule is not necesarry. Any underexposed shot (and -3 is VERY underexposed) doesn't provide additional clean information to the 0 and +3 shots.

My concept here is slightly different: take one shot making sure that you capture all highlights, but not underexposing at all, just make sure you don't blow information (a RGB splitted camera histogram is good enough to check this). This is the most important shot of all and after it you can forget about any additional underexposed shot.
If the camera has punctual measuring (don't know if this is the right term in English), the classical trick is measure the highlights and overexpose that measure by over +2EV according to the camera particular metering behaviour.
Afterwards comes the funny part: additional shots with RADICAL overexposures. 2, 3, 4 f-stops... The only care that has to be taken is not exceed a 4 f-stops gap between every two shots or some areas could start to suffer quantization problems or at least not reach the degree of tonal richness we are expecting from the whole image.

For the example of the couch I used:
0 EV, +4 EV
For the example of the blue room I used:
0 EV, +3 EV, +6 EV

Taking this last example I simply made sure not to burn pixels in the 0 EV (in fact I have just analysed and got real {696;501;344} RGB burnt levels which is absolutely nothing in a 8 Mpx sensor. Those few pixels surely were on the ceiling lightbulb.
And I didn't use 0, +4, +8 because a crowd came to the scene and I could not keep on shooting; but 0,+4,+8 would have been desirable for a better recovery in the 13th stop (see histogram).

It works.

In your case, I am sure with -3,0,+3 you will achieve a fantastic noise reduction and DR enhancement in the shadows, but I am not so sure if the DR in your image clearly represents the DR found in the scene. I mean: my shots are blended and developed in linear mode, and then obviously comes the conversion to the colour profile (sRGB, AdobeRGB,...) that will introduce a non linear gamma correction that is unavoidable. But I feel comfortable to know that right the previous step to this is pure linear data where 1 stop of DR exactly corresponds to 1 stop of DR in the scene since the sensor is workig in its linear range. I think you have more control this way on what's going on with your image than in a PS blending.

I've spent several hours over the past few days comparing Jonathan's 'split top layer blending method' with 2, 3 and 4 RAW images of the same scene loaded into HDR

The 'split top layer' method using just 2 images does not need an underexposed image. The bottom layer should just be an image correctly exposed for the highlights, which usually means when converting in ACR, approximately a minus 1 stop EC adjustment should be applied to 'recover' highlights.

My initial impression was that I was still getting a hint of the halo effect, but I now believe this was due to traces of silicon sealant around the edges of the window panes and/or inappropriate adjustments with Photoshop's Shadow/Highlight tool.

What I have noticed is that HDR in PSCS2 is not able to recover highlights well. If the lowest exposure is a full exposure to the right, the highlights will be slightly blown. In order to avoid this, I think it's necessary to include an underexposed image when using HDR.

Loading 16 bit TIF conversions into HDR seems to produce some pretty awful results.

May I have your RAW files to test them with my routine?

Some people have come to me with different HDR programs (like Photomatix) and after fiddling some time with them achieved similar results to pixel selection for blending. But usually HDR's tone mapping enforces local microcontrast keeping low the overall contrast, and this provides very unreal results; it's simply a different concept. I prefer not to alter local nor overall contrast and let the user the task to get the best from the noise free image in the way he likes best (contrast curves, zone edition, even HDR on other sofware,...).

Sending the files right now. Hope my connection is not interrupted. In the meantime, here's the comparison between an HDR result and the lowest exposure converted in ACR with minus 1 EC.

These are actually 'print screen' crops of 200% enlargements. No processing has been done except adjusting the WB slider in the HDR shot for maximum highlight recovery.

Click to view attachment

ps. Sorry about that water pipe in front of the view. It's manadatory in Nepal to always obstruct a fine view, if possible. Fortunately, there are still lots of pristine views they have not got around to spoiling yet .

jani,

There is an upgrade path from CS3 to CS3 Extended. I did just that today and I was way outside of the 30 day exchange time period. It is important that your original purchase of CS3 was in fact an upgrade from an earlier PS release and not a purchase of the retail program. Adobe sales is not completely familiar with this program but customer support aided me with their internal documentation that allows a customer who upgraded to CS3 and then would like to upgrade to CS3 extended the opportunity to receive full credit for their purchase and pay just the difference for CS3 Extended. Contact customer support and they should be able to guide you through the process.

Hope this helps.....

Victor

GLuijk I like very much your concept to work directly in the RAW
I am using blending options since a long time , or photomatix, but to do that directly in the RAW wil be the best

GLuijk,

When do you plan to release the new version of the program? It is very interesting!

Thanks,

In a way, yes, but in a way no. What is normally talked about is blending multiple exposures to extend dynamic range.

This seems more of a technique to use 2 exposures of a scene that falls within the dynamic range of the sensor but pushes it to the limit to reduce noise in the shadow areas.

Granted the way to accomplish this is virtually identical. Just most of us wouldn't bother doing this if our histogram shows we aren't clipping either shadows and highlights.

It has given me some food for thought however, and I can see now there are times doing this may be useful even if I'm not after an HDR image.

I tested this technique as a regular technique in a photographic indoor session last week, but in high constrast scenes as windows opened outdoor. The result has been very good: easy shooting (just regular shot plus one +4EV additional shot for every scene), easy edition (just a curve on which I set a mask not to blow windows facing outside), and natural result with no noise and expanded dynamic range.

Find here some examples (Canon EOS 350D + Canon 10-22 and a really bad tripod):





Taking the first scene: the 350D is too noisy to record outdoor highlights and indoor texture detail in a single shot. This happened when trying: highlights blown if texture is kept (picture left), texture is gone if highlights are kept (picture right):



Blending result:

What did you do to blend the exposures? You mentioned a mask, how did you make it? Jim

I used a little program I wrote to automate it, the download site must be somewhere in this thread.
The mask was just in the edition stage, to preserve the windows from blowing.

Yes, please let us know when the GUI version is available. Your work is outstanding!

John

Indeed! Your interiors are very esthetic, GLuijk. Sorry I haven't got around to sorting it out with my ISP about the errors I get when I try to send you 12mb file files.

I should (could) be on broadband, but I object to signing a 24 month contract when I intend travelling a lot in the near future.

why don't you put your raw files on a pendrive and try someone else's fast connection?

GLuijk,

my complements to your work on this new tool.

I am curious, whether you adjust the alignment of images while merging them together? There is "HDR alignment tool" which can curently do that, and probably could be used to provide input files to your program, yet not in the RAW format...

If you need any help with specific RAW files, I can offer Fuji S3 and Mamiya ZD files for testing.

Please let me know,

Thanks,

Hi, for all those that could be interested, I have just uploaded an English version of the article explaining this technique: ZERO NOISE PHOTOGRAPHY

Hopefully I will write a final version of the program ready to use along this month (Aug 2007).

Regards

Amazing tool

Looking forward to it!!!
Thanks

Thanks! The latest examples you provide are most impressive, no haloing or artificial/painterly look usually connected to "true" HDR images. I look forward to your program!

Any update on this?

Looks great. Charge a reasonable fee for a decent GUI based Windows program and I'll pay up

Quentin

Me too, for a Mac version.

Julian