Overexposed? How can I fix?

[Kerbouchard]

You really haven't 'recovered' any data. You have simply created new data to replace what was not there to begin with.

The 'data' is there. It is just in a different channel. If I would have had a bit more time, it would have been better balanced. Like I said, it's just a 2 minute edit.

As is, I would probably put my version over the original at about 60% opacity to create a final. With the original file, a lot more data could have been recovered(or as you prefer, recreated).

 

[480sparky]

You really haven't 'recovered' any data. You have simply created new data to replace what was not there to begin with.

The 'data' is there. It is just in a different channel.

So how do you make yellow with just the data in the blue channel?

 

[Kerbouchard]

You really haven't 'recovered' any data. You have simply created new data to replace what was not there to begin with.

The 'data' is there. It is just in a different channel.

So how do you make yellow with just the data in the blue channel?

By utilizing the luminosity blend mode that does not effect color. Basically, you treat each channel as an individual black and white image and blend the channels so that you can recreate the detail that you want.

Basically, duplicate the layer, adjust the channel that you want to bring out, and change the mode to luminosity.

 

[480sparky]

By the same token, you can do the same thing using paint on a print.

 

[cgipson1]

Since everybody said it couldn't be done, I figured I would give it a shot from the jpeg you posted. For fun, here is a two minute edit.

Color is way off......

 

[Kbarredo]

You can't repair this in photoshop. Parts of the flower are so bright that it is simply seen as white. so even if you lower the brightness it will still be white.

 

[Kerbouchard]

Since everybody said it couldn't be done, I figured I would give it a shot from the jpeg you posted. For fun, here is a two minute edit.

Color is way off......

It was a two minute edit... It wasn't meant to be a final result. It was just an example of the data being reconstructed.

 

[Bukitimah]

You shot this using f5.6, 1/250 sec and ISO 100. I think the light on that day is very strong, in the middle of the day? Using PP, I hope you would like this better

 

[Kbarredo]

Its a little better but some parts are still blown out. Try taking a picture with the flower in the shadow.

 

[Ysarex]

Since everybody said it couldn't be done, I figured I would give it a shot from the jpeg you posted. For fun, here is a two minute edit.

Excellent observation -- this technique Kerbouchard identified is legit with qualification. In this specific case it really couldn't be done since the blue channel is in fact clipped. However the blue channel is barely clipped -- it's hard to overexpose the blue channel with a yellow rose. To really have a case of overexposure where the only remedy is replacement data, you've got to blow all three channels. Most of the data that appears blown in this rose is in fact recorded in the blue channel. Therefore, as Kerbouchard correctly pointed out, the blue channel data can be used to reconstruct the missing data in the red and green channels. This process is not the same as manufacturing replacement data. "Photographic data" is quite unique and extremely difficult to mimic. The texture, exact lighting direction, noise pattern and shadows on the rose petals can't be painted in convincingly. If it's intact in the blue channel it can be transferred back into the red and green channels.

I've done this many times to save a sorry photographers butt. No offense to the OP who is learning here; this is the qualification I mentioned above. If you ever have to do this with one of your own photos you should be put over someone's knee and spanked. If you do it again you should have your camera taken away for a week. So we're talking about a method here that only applies when you've screwed up really bad and should be punished.

Hey Sparky, this works in GIMP, just for reference though since we all know you'd never need it . Create an empty layer above the background layer. Paste a copy of the blue channel into that empty layer and set the blend mode to Luminosity. The tone response of the blue layer is now the tone response of the image. The result is a disaster since you only want the blue channel tone in the blown highlights, not the rest of the photo -- you need a layer mask. You'll have to add a highlight layer mask to the blue channel. You've got tone now but you're still missing color where the original highlights blew to white. Create another empty layer above the blue channel layer and sample a highlight color from the background layer. Paint bucket that color into the empty layer and set the blend mode to Multiply. Again you'll need a layer mask to restrict the color to the highlights. In the illustration below I marked locations where the blue channel was clipped, the color there now is from my flat color layer -- you see it in the leaf. It's color without texture and as such is non-photographic -- beyond repair.

So, I really went to this trouble because if you look at this technique carefully and think about it, some really intriguing possibilities emerge. Think b&w and filters. I'm thinking about a photo by Ansel Adams of the San Fransisco Golden Gate before they built the bridge. The clouds in the sky are breathtaking. As a photographer you see that photo and say to yourself, yep that's old Ansel and his 25A red filter -- nothing like puffy white clouds in a blue sky shot b&w through a red filter. Now you've got a color photo with white clouds in a blue sky and the day was a bit hazy and wouldn't it be nice if those clouds popped like you know a good old b&w would with a red filter over the lens? You've got a red channel! Why not use it?

Joe

 

[Kerbouchard]

Since everybody said it couldn't be done, I figured I would give it a shot from the jpeg you posted. For fun, here is a two minute edit.

Excellent observation -- this technique Kerbouchard identified is legit with qualification. In this specific case it really couldn't be done since the blue channel is in fact clipped. However the blue channel is barely clipped -- it's hard to overexpose the blue channel with a yellow rose. To really have a case of overexposure where the only remedy is replacement data, you've got to blow all three channels. Most of the data that appears blown in this rose is in fact recorded in the blue channel. Therefore, as Kerbouchard correctly pointed out, the blue channel data can be used to reconstruct the missing data in the red and green channels. This process is not the same as manufacturing replacement data. "Photographic data" is quite unique and extremely difficult to mimic. The texture, exact lighting direction, noise pattern and shadows on the rose petals can't be painted in convincingly. If it's intact in the blue channel it can be transferred back into the red and green channels.

I've done this many times to save a sorry photographers butt. No offense to the OP who is learning here; this is the qualification I mentioned above. If you ever have to do this with one of your own photos you should be put over someone's knee and spanked. If you do it again you should have your camera taken away for a week. So we're talking about a method here that only applies when you've screwed up really bad and should be punished.

Hey Sparky, this works in GIMP, just for reference though since we all know you'd never need it . Create an empty layer above the background layer. Paste a copy of the blue channel into that empty layer and set the blend mode to Luminosity. The tone response of the blue layer is now the tone response of the image. The result is a disaster since you only want the blue channel tone in the blown highlights, not the rest of the photo -- you need a layer mask. You'll have to add a highlight layer mask to the blue channel. You've got tone now but you're still missing color where the original highlights blew to white. Create another empty layer above the blue channel layer and sample a highlight color from the background layer. Paint bucket that color into the empty layer and set the blend mode to Multiply. Again you'll need a layer mask to restrict the color to the highlights. In the illustration below I marked locations where the blue channel was clipped, the color there now is from my flat color layer -- you see it in the leaf. It's color without texture and as such is non-photographic -- beyond repair.

So, I really went to this trouble because if you look at this technique carefully and think about it, some really intriguing possibilities emerge. Think b&w and filters. I'm thinking about a photo by Ansel Adams of the San Fransisco Golden Gate before they built the bridge. The clouds in the sky are breathtaking. As a photographer you see that photo and say to yourself, yep that's old Ansel and his 25A red filter -- nothing like puffy white clouds in a blue sky shot b&w through a red filter. Now you've got a color photo with white clouds in a blue sky and the day was a bit hazy and wouldn't it be nice if those clouds popped like you know a good old b&w would with a red filter over the lens? You've got a red channel! Why not use it?

Joe

Awesome edit and explanation. You definitely put a lot more effort into it than I did and I just wanted to say it is appreciated. Nicely done.

And yes, the blue channel was clipped a bit in the JPEG posted. With the original, it could have probably been recovered.

For your edit, about the only thing I would have done different is with the two lower left petals(maybe three...the ones in the shade), I probably would have used a layer mask at about 40% Opacity to let the original come through to get it a little less 'muddy' and balance it out a bit better.

That was the problem I had with the edit and why I quit so soon. Seemed like most of the 'improvements' just made it look a bit more 'muddy'. Having the RAW file would help a ton. Create as a smart object and then reprocessing for each of the individual elements and then blending with layer masks would be the way to go. The Blue channel would still have to supplement the Red and Green, but it would be a lot easier to get natural colors and an overall balanced image.

In any case, again, nicely done.

 

[Ysarex]

snip...
Awesome edit and explanation. You definitely put a lot more effort into it than I did and I just wanted to say it is appreciated. Nicely done. snip...

Many thanks, I did want to pick up on this because of the tangential potential. I'm constantly using a monochrome blend layer (often one of the RGB channels, sometimes the L channel and sometimes a channel mixed derivative) to alter the color original's tone response. I'd be crippled without the technique. Here's a simple example -- caught the Ed Renshaw pushing up the Mississippi in central MO. The top version is textbook processed: white balance, Levels to tweak density and set end points and a slight contrast boost in Curves. The bottom version uses an extracted red channel to increase contrast with an Overlay blend. The difference in the two skies is dramatic and it derives from the tone response of the red channel. The clouds near the horizon that begin to disappear into the Midwest haze in the top version, stand out in the bottom version.

One more time going back to the OP's original photo and you're assertion that the clipped red and green channels could be reconstructed from the blue channel. I've had this argument before and I've taken your position; real overexposure requires all three channels blown. If data is retained in even one channel reconstruction is possible. My experience is that many camera JPEG engines with their "auto white mangle" engaged will clip one or two channels but not all three and so a repair is possible. This kind of stuff however bothers the BLEEP out of me when it's presented as a legitimate processing method (not saying you did that! -- others do). I think it's fair to say that I now devote 1/3 of my time in class to stopping my students from doing all the horrible things they've picked up on Youtube and Facebook. I have to detox them and constantly watch them for iPhoto and Picassa relapses. A good definition of stupidity would be anyone who gets into the habit of regularly repairing their screw-ups.

Ranting on with the theme and picking up on some of the other comments in this thread: My blood pressure quickly reaches dangerous levels every time I encounter the term "highlight recovery" :gah: I blame Adobe and now I find myself wasting days in class detoxing my students from Adobe drugs. This semester I had a student taking camera processed JPEGs straight into LR and using the highlight recovery slider to fix blown highlights -- if it sells more copies of LR!!

Joe

 

[TCampbell]

When in doubt about an exposure, learn to take a peek at the histogram. Not sure what camera you use, but usually when you're viewing a photo on the LCD screen of your camera there's generally a display button you can press repeatedly to change what sort of extra info gets displayed and one of those displays will show the histogram.

Here's the histogram for this image:



If you're not familiar with how to interpret a histogram there are numerous tutorials on the web. Here's one: Understanding Digital Camera Histograms: Tones and Contrast

You can think of the histogram as 256 columns ... standing side-by-side. The height of each "column" represents the number of pixels in your image that have that particular level of tonality. A tonality of 128 would be exactly in the middle. A tonality of 0 would be complete blackness. A tonality of 255 would be completely white (maximum values on all three color channels).

If the black graph is "climbing the walls" on either the left or right side then the image is clipped. If it's climbing the walls on the LEFT then it means you had so much data at tonality of 0 that REALLY you probably had data at tonalities in negative values that your camera simply could not capture. In other words, your "shadows" got clipped. If it's climbing the walls on the RIGHT then it means you had so much data at the maximum tonality of 255 that REALLY you probably had a lot of data at tonalities even higher than 255... but your camera couldn't capture those. As a result you lost data because your "highlights" got clipped.

Once you lose data there is no recovery (not unless you're a really talented artist and you can fake in something that make it look reasonable.)

A quick glance at the histogram after any shot where you question whether the exposure was good will tell you if you're safe or not.

ALSO... since you posted the exposure info (ISO 100, 1/250th @ f/5.6) we can see immediately that your exposure would have predictably over-exposed this shot. How could we possibly know that without being there when you took this shot and metered the light for ourselves? The "sunny 16" rule: The rule says that in "full sunlight" that a correct exposure should happen at f/16 (hence the "16" part of the name) AND with the shutter speed set to the inverse of the ISO speed. So at ISO 100, you'd use 1/100th (and for a lot of cameras the closest they can come to that is 1/125th).

We can see that this was taken in full-sun. You were at ISO 100. Had you been at f/16 and 1/100th, you'd have got a "correct" exposure (although the background would have been sharper than you may have wanted). Your shutter speed was 1 stop down (1/250th -- technically that's 1-1/4 stop faster, but close enough for round-off) but your aperture was 3 stops up. That means your exposure was predictably 2 full-stops (or 4 times more light) than you wanted for this shot.

If you wanted to shoot this at f/5.6 (as you did) to create the same level of background blur, then you're shooting 3 stops down from f/16 (f/11 -> f/8 -> f/5.6. That's three stops down.) So you have to move the shutter speed three stops UP. From 1/100 -> 1/200 -> 1/400 -> 1/800. And if you had a non-digital camera (mechanical shutter speeds were generally 1/125 -> 1/250th -> 1/500th -> 1/1000th) then 1/1000th would have worked. There would not be a significant difference in exposure results from 1/800th to 1/1000th. I know this "seems" like a large difference, but "stops" require a full doubling or halving. The actual difference would be only slightly noticeable -- close enough that either 1/800 or 1/1000 would have been useful.

The "sunny 16" rule was designed as a base-line back in the days when cameras didn't have built-in light meters. The rule is a bit more elaborate than what I've stated above (it has guidelines for what to do in light overcast, heavy overcast, various levels of shade, etc.) Even though modern cameras have built-in meters, knowing the manual rule can help as a kind of sanity check.

 

[markj]

great explanation, thanks

 

[KmH]

All of which points up the fact here's a hell of a lot of technical knowledge that has to be acquired to do digital photography consistantly well.