histogram question
mushy
Registered Users Posts: 643 Major grins
With a histogram like this, from shooting an osprey in flight against a bright sky what adjustments would you make to not have the histogram peak out quite so much?
May I take your picture?
0
Comments
In my opinion, there is absolutely nothing wrong with a peak in a histogram like this. The x-axis of the histogram is the tonal range (left is shadows, right is highlights). The y-axis is just a measure of how many pixels in your image fall in a particular tonal range.
So, a peak in the histogram just means that there are a lot of pixels in a narrow tonal range. That's just the composition of your image and is how nature made it. The three peaks in your histogram are all caused by the sky. If you look at the individual color channels, you will see that each color channel has a peak exactly where you see a peak in the overall histogram. The reason the peaks look large and are clipped on top is just because nearly 90% of the pixels in your image are from the sky. If Photoshop adjusted the scale so the peaks didn't look so tall and didn't show as clipped, then you would not be able to see anything else in the histogram because the scale would be so large. So, Photoshop assumes that all you really need to know is that there's a peak here so it adjusts the scale so you can see the other parts of the histogram (e.g. where the pixels from the bird are). You should know that in the y-direction a clipped peak in the histogram is not a bad thing. It's just a consequence of a concentration of a lot of pixels with the same tonal value and a scale that Photoshop chooses for your histogram. It does not represent any loss of information like clipping on the ends of the x-axis does.
If you want to experiment with spreading the peaks out to see if it improves your image, a common way for you do that is to use a curve. To spread the peaks out, you steepen the curve in the narrow tonal range where you have a peak. This will spread out the tones in that range and compress tones in other ranges where the curve is flatter (there's no free lunch with a curve, increasing contrast in one area always lowers contrast in another area). The ideal curve for contrast enhancement gets steeper in an area of interest and flatter in an area where you either have few pixels or where you don't care about contrast.
In this particular image, the sky is what it is and I don't think spreading the tones out in the sky is going to improve the image at all. In fact, when I try it on your image, all it does is show light fall-off in the corners and enhance noise in the sky, both undesirable characteristics.
This image does benefit from steepening the curve in the tonal area of the bird. This both spreads out the tones of the bird and brightens some of them to make the detail in the bird more visible. This curve steepens the curve in the tonal range of the bird and tries to keep the sky unchanged. The curve goes flat in a tonal range where there is almost nothing in the image so we don't see the loss of contrast in that range.
Here's the curve I used:
To get this changed image:
And here's your original image:
After doing this edit, the histogram looks like this:
The middle and right of the histogram is basically unchanged because the curve is unchanged in that area. The tones in the left of the histogram have been spread out across a wider tonal range. What used to be a small bump that represents the tones in the bird are now spread out across the whole left third of the histogram. The "comb-like" appearance of the left side of the histogram indicates that we are missing certain tones in the image. This is not generally a desirable characteristic, but it's a direct consequence of how steep we made the curve in this range to add contrast. While you don't generally want this characteristic, it's all a tradeoff because we also want more contrast in this tonal range, so you just go by how the image looks and let that be your guide. The image will print how it looks so it's not like we'll see posterization or something in a print that we don't see on screen. If you like the looks better, then it's better and we'll ignore this undesirable "combing" of the tones because the added contrast improves the image.
I hope this helps. Let me know if you have any other questions.
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
I could not have asked for a better or more detailed explanation of whats going on and what it all means. I am indebted to you!
Cool. Glad it was useful. There are a lot of misconceptions about histograms so I thought I'd try to clear a few of them up.
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
Thanks!
(And stellar job with bringing out the bird, too!)
www.tippiepics.com
Catapultam habeo. Nisi pecuniam omnem mihi dabis, ad caput tuum saxum immane mittam
http://www.mcneel.com/users/jb/foghorn/ill_shut_up.au
In the final histogram after the curves adjustment, something has clipped. I'll be it's just one channel & in this case is obviously not important as the image still looks good.
Another thing is the combing & spiking in the histogram; this will always happen when you mess with the curves & levels--it's just the nature of the beast. It's why we try and minimize our edits & remain subtle with them.
Ultimately the histogram is a guide, but your eye is the final test. If the image looks good with an ugly histogram, oh well. Then it is what it is.
http://www.chrislaudermilkphoto.com/
Photos that don't suck / 365 / Film & Lomography
Thanks jfriend - An explanation even I can understand.
Mushy
FYI - You might want to do a little housecleaning on your sensor. There seems to be some smoodge on the image.
Here's a great thread on sensor cleaning.
http://www.dgrin.com/showthread.php?t=8350
http://jburtphotos.com
http://jburtphotos.smugmug.com
Basic but makin' changes
Jamie
http://webwiz.smugmug.com