tone mapping - before and after

[pop]

Advertisement (gone after registration)

my opinion is that the tone-mapped version is the most dramatic and the "curves-adjusted" version is the most natural.

 

This is very difficult matter to decide.

 

I find your versions a bit on the dark side for my taste. Hence, I took the liberty of finding another curve to alter your baseline picture:

 

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

 

As you can see, this one is rather close to your tonemapped version in the shadows. While doing the curve, I noticed some missing values in the original curve. Is your baseline the unchanged original out of the camera?

[pop]

this is my first 'mapped' photo, so i'm pleased, but i hope you'll find it to be a good starting point.

 

rick - thank you for joining us. I think there are very interesting effects around the water's edge in you tonemapped picture. However, I like the sky better in your reference picture as it shows more detail close to the horizon. Perhaps you can somehow darken the sky a bit and give it more structure?

[pop]

I can not yet properly get my head around this Tone Mapping thing - however I did try to see if the original post would respond well to good old (?) fashioned PP.

Here is what I got from it - hope you don't mind 'Daveleo'

Can someone explain what advantage is gained from TM?

Osscat

Osscat - your version is indeed quite close to Dave's. The most marked difference I can see is in the sky. Yours is darker on the left hand side and brighter on the right hand side of the image, with a bit of apparent loss of detail in the brighter part. I suppose that could be dealt with, too, in traditional PP.

The algorithm of the script I introduced earlier in this thread has the property that the curves which adjust the output tones are not uniformly applied to all parts of the image. Rather, the curves are different for portions of the image with different brightness.

This leads to an overall reduction of contrast while raising the contrast locally. The cost is, of course, that areas which used to be coloured uniformly are not, any more. Rather, their tones undergo slight gradations. They are brightened close to dark areas and darkened close to brighter areas.

I rather suspect that a person experienced in "good old PP" has the tools at his disposal to do exactly that or some close simile. What this means is that Tone Mapping has been done before automated tools came into use.

Anyway, there's a bit of misunderstanding about. Technically speaking, tonemapping is just the process of adjusting the values of the source image to the gamut of the medium which is to present the image to the audience. Adjusting the brightness and contrast sliders is just as much an application of tonemapping as is the use of the colour rendition curve or the use of the shadow and highlight sliders which some products offer.

What's new here to some of us is that the tone mapping can be applied "unevenly" to different parts of the picture, and that there are algorithms which largely automate the process and which make it partly interactive.

[pop]

are HDR images welcomed here ?

Yes, they are. Welcome to the fun. As Dave already suggested, we'd appreciate seeing unprocessed versions along with the processed results, in order to see the influence of the process on the result. I know, we're a nosy bunch.

This looks like fun! (...) BTW, maybe this thread should move over to the Digital Post Processing sub-forum, where more people would expect to encounter this kind of info?

Kirk - it is much fun indeed. As for moving the thread to the PP subforum, I will refer to Dave's judgment. Unlike Dave, I am not shy about showcasing my ignorance in that place.

i THINK? tone-mapping is a poor-person's version of HDR

Not quite, actually.

HDR is MultipleCapture + MergingCaptures + ToneMapping (plus some fiddling between the major steps).

i am noticing that the effects on photos through glass are very dramatic(see Ece and Pop input above)

I don't think it's to do with the glass. It's rather exterior (sunlit) and interior (shaded) parts which make the difference. In Ece's last example you see the underside of the roof as well as what's inside the glass dome after she's done with the tonemapping.

That appears to be the point of the whole exercise. In a picture where either the bright parts get burnt on the monitor (or the print) or the dark parts drown in blacks, you could lower the contrast, but then the whole image becomes hazy. (The main optical effect of haze appears to be lowered contrast). Enter non-uniform tonemapping where you bend the tones of the whole scene such that both bright and dark areas are shown with details.

Kirk . . . i was so intimidated by the "digital PP" forum that i posted this here.. . . those guys are way way above my skill level and they also use Photoshop and LR/ACR (which i don't even know what that is !)

As I said above, I don't think we need to be intimidated. I dare anyone to sneer at the bit of fun we're having here.

Beyond that, Philipp is the expert to tell you the detail.

Ece, that's an interesting shot and an interesting observation about the repeated application of the process.

Nice sentiment towards the wrong person. I am no kind of expert on any kind of PP, least of all tonemapping. Just fumbling along and having fun.

Apologies BTW for multiple lengthy posts. My two other jobs are keeping me busy right now, and so I'm trying to catch up with things whenever I can. As one Goethe once wrote: "I am short of time at the moment; hence, I am sending you a longer letter than usual".

[pop]

I just had a look at another open source PP software, Krita.

I already knew that it did colour depths of more than 8 bits per channel. I just found out that it has several (not only one) tone mapping filters built in.

On the down side, the version which came with my OS is not as robust as I would like. I can not recommend it for production work or for irascible computer users, (except for those who feel they need a new computer, anyway). You have been warned.

[daveleo]

hey kids, it's okay with me to move it to the digital PP forum . . . i will request that.

[allamande]

Advertisement (gone after registration)

Warnings don't have much use, when you have an opening sentence like this:

 

[schmal]

I just had a look at another open source PP software, Krita.

 

 

Don't get me into further trouble, please

 

Ece

[rickp13]

rick - thank you for joining us. I think there are very interesting effects around the water's edge in you tonemapped picture. However, I like the sky better in your reference picture as it shows more detail close to the horizon. Perhaps you can somehow darken the sky a bit and give it more structure?

 

thanks, philipp

 

let's see if this is going in the right direction - a touch lighter in the landscape and a touch darker with emphasized clouds in the sky area.

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[daveleo]

Warnings don't have much use, when you have an opening sentence like this:

Don't get me into further trouble, please

Ece

 

that's kind of why i would like to understand what's going on behind the buttons and sliders . . . we can keep getting new software OR we can understand the process and use whatever software we have to get the effects we like.

 

i think . . . maybe . . ?

[pop]

thanks, philipp

 

let's see if this is going in the right direction - a touch lighter in the landscape and a touch darker with emphasized clouds in the sky area.

 

Closer, but not quite there, I would think. Somehow I feel the sky looks a bit "dirty" now. Since I am only a beginner with all that post processing, I don't see how I can help you there. :confused:2

[pop]

There's another study, this time a triplet. Left: the original shot of some elderly building., with the range of value stretched so as to include values from 0 to 255. Middle: processed in the GIMP using the AdvancedToneMapping script. Right: Manual non-monotonous adjustment of the brightness curve.

 

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

 

Again, it's a study of the technique involved, not a candidate for the MOMA.

[sanyasi]

For those interested in getting into HRD and tone mapping in more detail, I recommend a book by David Nightgale called Practical HDR. Here is the listing at Amazon: Amazon.com: Practical HDR: A complete guide to creating High Dynamic Range images with your Digital SLR (9780240812496): David Nightingale: Books I have looked at a number of books on the subject and this is the best in my opinion.

 

Nightgale explains the process and each control in detail for Photomatix, another program, which I forget the name of, and Photoshop. He is clear and to the point. While some view this as frivolous/over computerized processing, there is some strong theory behind it. Nightgale shows how to use Photomatix in two ways: one for creative, painterly photos and the other to expand the dynamic range portrayed in the finished photo, but remain truer to the reality that was captured.

 

I have had fun with the process, testing it for night photography. I have found, in many cases, that I prefer one of the unHDRed photos of the seven that I capture with the autobracketing on the Leica M9.

[pop]

A few words are in order to help us understand what the AdvancedToneMapping does. The step numbers mentioned in this post refer to the numbers in the post above.

The core of the function lies, of course, in steps 9 and 10. In order to understand those two steps, we have to understand how the GIMP works with layers.

In this special case, we have to look at the situation where exactly two layers are involved (no more nor less). The upper layer will be called the "Mask", M. The lower layer will be called the "Image", I. For each pixel to be shown on the screen, the GIMP will take one pixel from the layer M, the corresponding pixel from the layer I (the one hidden by the pixel in M). It will apply a mathematical function to the pair of pixels (m, i). The function depends on the "Mode" you selected for the upper layer M. The function can be as simple as an addition. It can be quite involved.

In the case of the "Mode" selected in step 9, the function is called SoftLight (or, rather SoftEdge).

What this function does is that it applies a different gradation curve to the image, depending on the value of the corresponding pixel in the mask layer.

The unity gradation would not change the picture at all:

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

If I understand the formula provided for the function given for SoftEdge correctly, the gradation curve applied to the image for any given value of M would like something like this:



Since the result of the formula varies for each value of M, we get a sheaf of gradation curves, one for each possible value of M:



What this appears to mean is that in steps 9 and 10 the picture will be brightest where the mask is darkest. It also appears to mean that places of the picture which - in the original - are very bright or very dark will be rendered darker than those of medium brightness.

From those steps working backward, we find steps 4 and 5 refreshingly simple to understand: we produce a B/W negative of the picture to be tonemapped.

Step 6 blurs the negative. The Gaussian blur recomputes each pixel such that it has the average value of all the pixels within the blur radius. The result is that sufficiently large areas will keep their colours while details will become less visible and sharp edges become gentle slopes. Since the blurred negative will be used as the mask layer in steps 9 and 10, we can see how the angle of the slopes will affect the final result.

Caveat: I am altogether uncertain whether I have interpreted the formulas for the SoftEdge mode correctly. Hence, my curves could be wildly off.

I wonder if I am making sense to anyone here.

Lest anyone think I invented all that nice background: I didn't, I just took it off the GIMPs documentation page and wrote here what I probably misunderstood, it being somewhat late in the day. Perhaps some engineering type members could look at all those lovely formulas.

[pop]

In order to understand those two steps, we have to understand how the GIMP works with layers.

 

Judging by the lively discussion after this first attempt, I presume that it did not win me a cigar. Small matter as I don't smoke, anymore.

 

To better study what the different rules do when combining layers, I designed a fiendishly intricate image and an equally intricate mask. Here are both the image and the mask.

 

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

 

 

I then imported both into the same (empty) GIMP picture using the menu command File/Open as layer (or similar, I'm translating from German). Now I can change both the mode and the opacity for the first (upper) layer while closely observing what the brightness of each square in the resulting image does:

 

 

In the original GIMP image, the strips are uniformly coloured without any borders at all. However, I am not sure about the compressed version I have been able to upload. But even in the original version the eye insists in seeing borders between the areas and even varying brightness within a single strip or square. Be assured that it's all in your head.

 

If someone would prefer to use the original images I used in the GIMP, feel free to PM me so I can mail them to you.

 

The values of the stripes in hexadecimal notation, from darkest to brightest: 000000, 101010, 202020, 303030, 404040, 505050, 606060, 707070, 808080, 909090, A0A0A0, B0B0B0, C0C0C0, D0D0D0, E0E0E0, F0F0F0, FFFFFF (17 stripes in all).

[daveleo]

Philipp . . . . wait . . . .I had erased my subscription to this page . . . . don't get disheartened, I for one am trying now to catch up with you.

 

Thanks for all this info . . . after I do my homework, I'll be back

[daveleo]

Philipp

I believe you about the smooth vs stepped gradients . . . I took a screen cap (no jpg compression), and it is below. . .

 

I also realize now that the magic is in the math algorythm(s), which is probably more than I desire to know. However, the general process is becoming more clear, and I see some wild results choosing other Modes for the layers (which calls in other algorythms) . . . . which is probably a book in itself.

 

Am going to do step by step manually for a while and see how much "art" I can create (we are getting beyond "photography" I think.

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[pop]

If I understand the formula provided for the function given for SoftEdge correctly, the gradation curve applied to the image for any given value of M would like something like this:

 

[ATTACH]213551[/ATTACH]

 

 

My subsequent post clearly shows that this curve is nowhere near what actually happens.Clearly, I don't grok the formulas in the GIMP's documentation of those modes.

I advise anyone to ignore my interpretation of that mode I gave above. Rather, look at the transformed grey scales.

[daveleo]

My subsequent post clearly shows that this curve is nowhere near what actually happens.Clearly, I don't grok the formulas in the GIMP's documentation of those modes.

I advise anyone to ignore my interpretation of that mode I gave above. Rather, look at the transformed grey scales.

 

I agree . . . it is not that simple . . . look at the screen cap from the documentation . . . that Rs term is in there, plus there is "I" (not just "M") . . . the graph shouldn't be as simple as you showed it.

 

Also . . . as I am retired, will not plunge deeper into the math details . . . I am convinced !

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[pop]

This time, I took quite a different route. I applied the "mask" (the grey chart with the horizontal stripes) to the "image" (with the vertical stripes). I then measured the brightness of each of the small squares in the array.From the measurements, I drew a graph for each individual stripe of the mask. This gives the following sheaf of graphs:

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

What it means is this:

For medium values of the mask around 128, the mask does not have a noticeable effect on the image. The graph which shows output value vs original value of the image is pretty close to the "normal" diagonal line which means identity.

For a dark mask - i.e. a mask with values smaller than 128 - the curve is flatter for low values in the image and steeper for higher values. The contrast is lowered for the darker parts of the image and increased for the brighter parts.

The inverse happens for a bright mask. The contrast is increased for the darker parts of the image and lowered for the brighter parts.

Now it is necessary to remember that the "image" in this operation is the original image we wanted tone mapped. The "mask" - the upper layer - is (a) the negative of the original image and ( unsharp.

Hence, for a spot in the image which is on the average very bright, our mask will be very dark. In that very bright part the contrast will be increased for the brighter spots and lowered for dark places. Since the mask is unsharp, the effect will be strongest at the brightest spot. It will gradually fade the more distant we are from the brightest spot.

This is just the bare bone of an explanation. It does - for example - not take into account that the mask has been partly but not completely de-saturated.

Discuss.

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[daveleo]

don't rush me . . . i'm thinking . . .