Histogram Confusion!

[Astrosurf]

Can someone explain what I see when I look at a histogram of my images? I've tried to understand MartinB's tutorials, but I don't get all of it! I can't understand how the shape equates to the image. I can see it change after I've tweaked it in curves, but that's all!

Here are screen shots of an unprocessed image of M42 and then two processing attempts.

Alexxx

[ollypenrice]

On the x axis, dark pixels on the left, bright pixels on the right. On the y axis, the number of pixels shown for each level of brightness. Higher is more.

On an astrophoto there are usually lots of dark pixels for the background sky so the peak is towards the left. The long flat line to its right are the stars, mainly.

Now the key bit, the stuff to the left of the main peak. Some of this is faint data and some is noise. So some you want to lose but not all.

There must ALWAYS be some flat line to the left of the main peak or you have clipped faint data. It cannot be retrieved other than by starting again. This applies to the second two screenshots. Naughty!!! Classic black clipping.

The first shot does not look clipped but you have a double peak. This is a colour or RGB image? Most likely if you split the channels in the histogram the peaks for each colour will be in different places. This needs attention.

You can't move the peak to the right so the colour which is out of balance can only be moved to the left by clipping back on the left. The top left of each colour peak should line up (unless this obviously looks wrong, but I doubt it will.)

Avoid clipping the black point, you throw away data and end up with a black sky.

Olly

[Astrosurf]

I think I'm going to have to have a one-to-one image processing lesson!

[ollypenrice]

Was my post at all useful? I can always try to rephrase it or clarify.

Olly

[Astrosurf]

Sorry Olly. I hope you don't think I was being dismissive. I just don't get everything people say and so I didn't want to bother you guys any more as I see myself as a hopeless case! I need to be shown really. Most of what you said I understood though, and it was very helpful. I wasn't sure exactly what you meant by:

The first shot does not look clipped but you have a double peak. This is a colour or RGB image? Most likely if you split the channels in the histogramthe peaks for each colour will be in different places. This needs attention.

You can't move the peak to the right so the colour which is out of balance can only be moved to the left by clipping back on the left. The top left of each colour peak should line up (unless this obviously looks wrong, but I doubt it will.)

Especially those bits underlined. I wondered if clipping back on the left means move the left-hand slider on the histo to the right a bit, but I thought I mustn't lose that data? I'm not sure how to avoid the problems. The image was a simple colour TIF.

[MG1]

Don't be daft....how do you think we all learnt anything...

Your image is mad up of data in the Red, Green and Blue Channels. The combination of data in these three channels contribute to creating all of the colours you see.

This creates what is refered to as an RGB image.

If you go to the Image>mode menu you will see RGB colour. There are also other structures for colour files (eg CMYK)

You will also see the bit depth there. The greater the bit depth the wider range of colour values.

a 'bit' of info here: Bit-Depth

histograms: Free Photoshop Guides Blog Archive Understanding Histograms by Example

somtimes it's easy to assume knowledge when folk try to explain things, in an atempt not to be patronising in any way...so don't be affraid to ask, or google for an answer then ask for clarity...

it's a steep learning curve, but you're already well on the way...you've got some great data.

Michael

[ollypenrice]

OK, to try to clarify: you can ask Ps to show you the histogram for each colour in turn. I think they changed the shortcuts but it used to be Ctrl1, Ctrl2, Ctrl3 for the R, G and B channels.

In the first screenshot they are all overlayed but there is a split in the peak - a little crack starting at the top and going down a way. I guess this just means your R, G and B are not lined up, which is common.

In each channel moving the slider to the right does indeed cut back and make the peak move to the left. You do this to get the colours to line up. Top left of each peak should align.

Will this clip data? Maybe, but you have to look hard at the faint stuff in the picture. Some of the flat line to the left of the peak is noise and wants clipping!

A little bit of flat to the left of the Peak is healthy.

A good rule of thumb; open Curves (Ctrl M) and put the cursor on some plain background sky and Alt Click. A point will pop up on the curve graph showing how far up your background sky is. 15 is a good value. Less and you have probably clipped data and the sky will also look artificially black. If this is the case go back in History and clip back less.

I try to leave a value of 20 or so in the background sky until the very end of the work. Somehow I always seem to nibble away unintentionally at that background sky and end up with it too black, hence keeping 20 as the working minimum till the end.

Olly

[roundycat]

Olly has made a good job of explaining what is clearly a tricky concept for a lot of people to grasp. One thing I would add to all this. As you make the adjustments concentrate on what they are doing to the picture. As has been said many times before, it is the picture you hang on the wall, not a screen grab of the histogram!

Dennis

[RobH]

Alexx.....have a look on my website (address in my sig) under 'primers and tutorials'.

There's a section all about the histogram witrh different examples to look at.

Cheers

Rob

[Astrosurf]

Hey, that's just great guys! And thanks for your Webpage Rob. There's nothing like seeing it in images. I learn so much better that way. So much of what people have said has started to come together.

Now to try it out . . . !

[NickK]

A Histogram is a tally of the colour values in the picture.

The number of values a picture pixel is given by it's colour depth. This is where you see 16 bits for CCDs mentioned. If the CCD is monochrome then each pixel will have a value between 0 and 65,535 (ie 2^16-1).

So if it's complete black with a 10x10 picture you'd get a tally of 100 at the 0 colour value. If you had a CCD image with half black and 50% grey (value 32,000) then you see two spikes - one spike of 50 at value 0 and one spike of 50 at value 32,000. Having 100% white would be a value of 65,535.

The curves change the scaling of those values. So moving the left most marker down to halfway to the right means the computer will now see the value of 32,768 (rounded up) as the new 100% white colour. All other values are now scaled to fit.

I could go on but as you have it already I'll not bore you (or confuse the matter) further.

I wrote my own software with histograms, they're very simple one you get the knack. However I would only use Histogram curves to flatten the data to bring details out on screen as the very last step in processing. The reason is that compressing can loose information (which reduces quality when stacking for example).

[roundycat]

Nick, your last sentence seems to imply that you sometimes adjust before stacking. Surely I haven't got that right?

Dennis

[NickK]

Nick, your last sentence seems to imply that you sometimes adjust before stacking. Surely I haven't got that right?

Dennis

Clarification - that the reason why you shouldn't do it before stacking.

[Astrosurf]

Thanks so much guys. I'm working through all this at the mo!

[red dwalf]

me too !

i`ve learnt something tonight, or learning something, going back to my older images and having a play around with them.

[Astrosurf]

SGL's just so brilliant isn't it?!