Burning and dodging

[cathodeRay]

I'm working on restoring an old B&W photograph. One of the more awkward blemishes is the bottom RH corner has a large wedge of white across it. Though it could be real (the aft edge of the cabin top), I think it is damage from the original developing and printing (or maybe it didn't get fixed, as in chemical fix?). Anyway, I've tried to patch (mostly using the patch tool, which has recently been in a better mood than it sometimes is and yes I know it's not an 'official' tool but so far it has done the best job) the area with some minor success - see below (the diagonal line of the original photo/blemish boundary can still be made out - but what would be really helpful to help smooth out the patching is to be able to dodge and burn with custom control of what constitutes lights, shadows and midtones, ie target the dodging/burning very precisely eg so already dark enough areas don't get burnt to dark. I suppose it could be done at high zoom with a small brush but then mottling starts creeping in. I prefer a slightly large brush with a soft edge. I could also do lassos (selections) for each greyscale level but a gain very tedious as the degree of precision needed would be considerable with multiple lassos. Instead, being able to 'dial in' "Only burn pixels between 30 and 60" (as on 256 shades of grey) or set say the cut off between lights and midtones at X value would be very helpful. The much smaller thinner black wedge isn't a problem if the white can be fixed. Cropping off the white wedge isn't really an option, in that other significant parts of the photo would also be cropped (and also the composition of the whole will suffer).

Any way of achieving this? Or another 'mega-solution', perhaps using a different technique?

BTB and in passing, I cannot find a way to have the crop tool auto-jump to the edges of an image (as when cropping to the image below). I normally (in other programmes - whoops, and not its the one that shall not be named) just drag the crop rectangle from there I want the crop to start to off the corner and it automatically jumps back to the edge of the image. In PL it appears to include the non-image part of the rectangle as part of the crop, and I have to meticulously replace the crop line exactly on the image edge.

Many thanks for any help in advance.

restoration.jpg

cathodeRay

[bkh]

BTB and in passing, I cannot find a way to have the crop tool auto-jump to the edges of an image (as when cropping to the image below). I normally (in other programmes - whoops, and not its the one that shall not be named) just drag the crop rectangle from there I want the crop to start to off the corner and it automatically jumps back to the edge of the image. In PL it appears to include the non-image part of the rectangle as part of the crop, and I have to meticulously replace the crop line exactly on the image edge.

If you turn on auto guides, the crop rectangle will snap to the image borders. Depending on the settings in Preferences -> Display -> Guides you can temporarily enable auto guides by holding the Alt key.

Alternatively, use "Select All" to select the entire image, then shrink the crop area (hold hown Shift if you don't want to keep the aspect ratio).

Btw., it's often useful that you can crop outside the image, especially if you use the crop tool to straighten or unskew images.

Cheers

Burkhard.

[Hoogo]

What about splitting the frequencies? It looks to me as if smudge and clone brush could do a lot on the low frequency layer.

[photoken]

Without seeing the unrestored section, it's hard to suggest another approach....

As it is, the Darken tool can be used to fix the areas above the white wedge, but it will be tedious -- zoom waaaaay in and use a very small brush, targetting either the midtones or the lights. Vary the brush size and intensity to essentially re-paint the areas.

You can also try adding multiple adjustment layers such as the Curves and modify the curve to target specific tonal ranges. Then create masks for the adjustment layers to limit their effect to specific areas. I don't think this will be any less tedious or time-consuming, however.

You might have hit on the best solution -- accept the white wedge as being the cabin top and maybe paint some texture, etc., on it to make it look more like the cabin top.

[cathodeRay]

Thanks all for your replies. PL never ceases to amaze with the tricks it has hidden behind various keys - Ctrl+A with the crop tool is obvious when you think about it, Shift + Drag to turn off keep aspect ratio less so but solves the problem - a quick easy crop.

Ken - I attach the same part of the original scan. The fuzzy border between the white wedge and the image proper could be because it's the out of focus aft edge of the cabin top (coachroof for salty types) but (a) I would expect to see a companionway (access to the cabin) hatch there, even if blurred, and (b) the rest of the image has great detail and is in such almost perfect focus - just look at that pinstripe! - that I find it unlikely that the aft edge of the coachroof should be so out of focus - so I think it is light (or shadow) damage of some sort. The photo appears to have been taken from the side deck, possibly with some sort of tripod to keep things steady - the date is probably late 1800s, possibly 1890's.

Hoogo - I'm still getting to grips with what frequency separations can do. The primary task here is to clone/patch the white area with something that looks OK (= doesn't ruin the photo/make the viewer think what on earth is that about). Careful work with the patch tool got me to what you see in the original post image - there's something there, but it's not perfect. Selecting (lasso) the whole area and doing Remove Object (PL's content aware fill method as I understand it) produced bizarre results, as did other external content aware fill programmes like InPaint. It's not difficult to see why. And then there's the problem of matching up the pinstripes. Local patching allows a reasonable stab at doing that (repeated clicking of different source points until it's 'good enough').

At this stage, I am keen to see if I can reasonably easily get the tone (by which I mean overall darkness/lightness) of the patched area right. I've tried most of the things I can think of (eg luminosity copy/healing/darken (burn) brush) but the results are, ahem, patchy, with the need being, as I see it, to darken areas that need to be darker, leaving already dark enough areas as they are - hence the interest in an ability to set custom lights/midtones/shadows ranges. I do suspect this could sort of be achieved by setting masks (lassos) as needed but it could all get horribly complicated horribly quickly (which mask is which, which is on/off etc, some overlapping). Another approach might be some sort of 'brush if' setting, but I've yet to see how to implement that.

Any thoughts very welcome.

original_scan.jpg

cathodeRay

[cathodeRay]

After some further experiments on the part restored image, it occurred to me the (background) image could in fact almost be the (curves) adjustment layer mask needed: darker areas protected, light areas more affected. There seems to be a bit of almost voodoo in using an image as its own mask, but actually, I guess what's happening is it is acting as a luminosity mask, for adjustments to itself. So only a bit voodoo-ey. Improvement not good enough to post here yet but a step in the right direction.

I've also discovered the adjustment layer colour filter dialog provides the range limiting sliders I had in mind earlier (I had been thinking colour filter might be a colour (only) filter). Applied to an adjustment layer, my understanding is that this is a 'blend if' type of method. But I have still to work out how to set the range limit. With the colour filter dialog open, placing the cursor over the image gets an eyedropper, but it seems it's just for info, no click to add the value at that point to the sliders. I'm also trying to figure out what the manual says about what affects what... At the moment, the sliders seems to create very hard edged effects, at the border between affected/not-affected pixels.

cathodeRay

[photoken]

I think it is light (or shadow) damage of some sort.

Yes. Seeing the original, I agree with your reasoning.

Since there's nothing at all in the white wedge, you'll be drawing new image information in there anyway, so I'd look for an image of an awning, or an image of a wood railing and simply paste that into the area. Blend the patched image, soften it a little bit, and call it a day....

[Hoogo]

Hoogo - I'm still getting to grips with what frequency separations can do. The primary task here is to clone/patch the white area with something that looks OK...

I have mistaken that white cloudy areas as the parts of the picture you would like to repair. I have not seen it as the first steps already done to repair it

At this stage, I am keen to see if I can reasonably easily get the tone (by which I mean overall darkness/lightness) of the patched area right. I've tried most of the things I can think of (eg luminosity copy/healing/darken (burn) brush) but the results are, ahem, patchy...

The frequency separation did work quite well. Your patches already made nice details, and after spliting I used the smudge brush and worked on the low freq layer to make the overall brightness.

[cathodeRay]

Hoogo - that's pretty impressive! In fact I was thinking (borrowing a bit from Ken's idea) of adding a gradient 'shadow' from the bottom RH corner to cover the worked on area as well, which will further help decrease any distraction. Very smart observers might note the sun/shadow positions aren't entirely in sync (a bit like a painting of a sailing ship where the flags are blowing the wrong way given how the sails are set - but all no doubt in the interests of artistic merit) but it is not entirely impossible to imagine that that area might have been in shadow.

On a related point: as you can see I did some basic corrections (and a lot of local spot repairing) before attempting to fix the wedge, ie the patching is more at the end of the workflow. This was because I figured I would have better patch sources to work with, but I wonder if there is a case for doing the major repairs before doing anything else, so the major repairs get sucked into (and so possibly more blended into) the general corrections as well? Just a BTW sort of question.

cathodeRay

[photoken]

In fact I was thinking (borrowing a bit from Ken's idea) of adding a gradient 'shadow' from the bottom RH corner to cover the worked on area as well, which will further help decrease any distraction.

I spent some time looking on the Internet for appropriate patches (wood railings, wood decking, canvas, etc.), but didn't find anything that would work. Something like the corner of a canvas tarp -- just something that would give some definition and texture to that area before de-emphasizing it....

This was because I figured I would have better patch sources to work with, but I wonder if there is a case for doing the major repairs before doing anything else, so the major repairs get sucked into (and so possibly more blended into) the general corrections as well? Just a BTW sort of question.

Like you said, if I'll need some good patch sources then I'll do the fiddly repairs first. I tend to work from the general to the specific, so usually I'll do the major work first because that makes it easier to keep the "big picture" consistent.

[cathodeRay]

Well, after going back to the original scan and starting all over again from the original scan, I think I've got a good enough fill for the wedge area. Used just about all the techniques mentioned, so thanks to those who generously made them, including a roll my own frequency separation which means I now think I understand the basics (for now, but I am probably just about to blow it). On these layers (I did just a low frequency/high frequency pair), I think I read somewhere here (or maybe elsewhere) that some other form of blurring for the LF layer is now considered a better option than a Gaussian blur, which is what I used - but I am not sure what it is? I also rather liked the logic behind using Image > Add Image > Add Inverted (or for 8 bit Subtract, and both are in effect subtractions) method to create the HF layer (a procedure available in certain other SW) but so far as I know there is no equivalent procedure or workaround available in PL. The appeal of the logic is that you subtract the LF data and then add it back in the final Linear Light blend, so in effect getting back all your data.

BTB, I also noticed when cloning (with the copy brush) that while the incredibly useful preview works with the soft edge brush settings I had selected, once I started dragging the brush settings became much harsher, ie almost no soft edge effect at all. Dare I wonder if this is some sort of bug? Or is it by design, or am I just doing something wrong?

Anyway, for what it's worth, here is the 'good enough' result. Any comments/thoughts very welcome (feel free to say it isn't 'good enough' if you don't think it is!):

restoration_2.jpg

cathodeRay

[Hoogo]

I did some experiments with frequency separation, too. I forgot if the attached action was really finished, but it works. The idea was to use adjustment layers for the whole thing, so you can change the type of blur or size afterwards, or a denoise plugin, or use the layer stack for other purposes where 2 pictures should add to one.

"Adaptive soften" or denoise plugins are often used instead of gaussian blur. Gaussian is perfect for mathematicians, but the others are more for practical users . Its easy to work on a "low pass" layer that lacks details, but has still some normal sharp edges.

Btw, great result.

[photoken]

Yes, great result.

The only quibble is that the coat looks a bit strange -- long like an overcoat rather than shorter like a suit coat....

[cathodeRay]

Thanks for the positive feedback - much appreciated. Hoogo - I used an essentially non-destructive approach as well, though I did edit the separation layers directly. Making all editing non-destructive is the ideal, as I found out (yet again) when I wanted to go back on an edit, but couldn't. Frequency separation as a technique is certainly impressive. My earlier understanding was that it was a complex recipe that had to be followed to the letter whereas now I see it more as an approach, with considerable flexibility in how you go about achieving the key step of separating the frequencies. It is still a bit counter-intuitive at least to me to think of image structure (light itself obviously has a wavelength, at least in classical physics) in frequency terms but I suppose it is a bit like ploughed fields on a hilly landscape. More importantly, once you've got the basics sorted out, it works!

Ken - I agree about the coat, but I think it is for real, at least it was on the day, even if added to today's photo. If you look at the original scan I posted earlier, you can see the coat extends far lower than a modern suit jacket would on the LH side where the wedge is less intrusive. I think it was probably a sort of morning or frock coat. Given the wing collar, monocle and watch chain, it seems they really did dress up to go sailing in those days. I have a large Beken print of Charles Nicholson in 1930 helming a large yacht (Candida?) in what looks like a stiff breeze (there's a thumbnail here). The lee rail is all but under but there he is standing on deck in a jacket (albeit 'only' a blazer) and tie in pressed trousers with turn-ups and what look like brogues, with absolutely nothing between him and the sea should he lose his footing!

cathodeRay

[Hoogo]

...whereas now I see it more as an approach, with considerable flexibility in how you go about achieving the key step of separating the frequencies. It is still a bit counter-intuitive at least to me to think of image structure (light itself obviously has a wavelength, at least in classical physics) in frequency terms...

My memory hook is what I've learned in physics about sound and wave forms. There's only the sinus wave, but you can add more and more waves of different length and amplitude to create any wave form. Same can be done in 2D with brightness instead of amplitude, and blurryness is like the frequency. Simple horizontal and vertical waves as used in jpg are already sufficent to sum up to anything.

Some funny thing in a similar area: Geocentric world view used cycles added to cycles to describe the movement of planets. What can be done when you add even more cycles?