Test Image with Uniform Histogram

[Photon Hacker]

The reason I am looking for this is to try and get an image that I can use to compare the effects on the histogram of different adjustments in PS. eg Shadows/Highlights vs Curves.

Or am I barking up the wrong tree ?

First, let me note that what we're actually talking about is the distribution of pixel values (For whatever it is defined to be, usually the average of the RGB channels whose meaning also varies). An histogram is a common representation of this distribution but they're not the same. To talk about the "effects on the histogram" in this context is like talking about the "effects on the speedometer" in a context regarding different types of tires (What we would then be interested in is speed, responsiveness, etc... not its representation). Unfortunately it seems like no other fellow forum use has noticed this mistake. Maybe this don't means too much to you but as an enthusiast of mathematics and formal logic I felt compelled to point this issue.

I think You're addressing the incorrect issue. What we are interested in, regarding photography is our perception of things, our perception of "different adjustments" in this case; with this in mind I'd say pixel value distributions are of little help here (Remember the black cat next to a coal pile and the white dog in snow).

If you want your comparison of "adjustments" to be meaningful for real photographs then common sense says you should perform the comparisons with real photographs, and compare how they look, not a pixel value histogram looks.

If you want to know more about the algorithms behind the "adjustments" then I'd normally suggest to look in the source code, however, the image manipulation program you inquired about is proprietary software; by design it prevents its users and general public to consult the details about its workings. I use and suggest free software like The GIMP and RawTherapee which allow and promote this and other rights.

Regards and good luck.

[Colin Southern]

I think the difficulty might be making sure you have a perfectly linear gradient. If pure white or pure black ends of the gradient are more than a pixel wide, the histogram might show up as having 'tails' rather than being flat. I'm fairly sure that Photoshop has a function that lets you split a color file into r, g, b channels. You might have to replace the individual r, g, b channels with linear gradients and then re-combine the channels. Or something like that. Good luck.

As I see it - if the histogram can show 256 levels - and we're applying a linear gradient - then we'd better make sure that the width of that gradient is an exact multiple of 256, or the histogram is going to have spikes or gaps in it as it tries to fit a square peg into a round hole.

Also - would it not be easier to just do it on a greyscale image?

[John C]

You are probably right about using multiples of 256. I'm sure that's why I subconsciously used an image 1024 pixels wide. However, your comments gave me an idea that I just tried out. I used 4 cycles of gradient within the 1024 pixel wide image (rather than the 1 cycle I used before) and I still get a fairly flat histogram, maybe a little better.

As for using grayscale, it just seemed better to have a full-color r,b,g file to test out. I'm really not sure what the file would be used for other than to view the results of brightness/darkness adjustments such as conducted using curve, histogram and the like. For me, creating a file with linear histogram is an interesting exercise. When editing photos, I am more interested in trying to achieve a particular result and not so much the details of what the program is doing.

[dje]

Thanks everbody for your latest comments. I think this is all getting too hard - my head is spinning !! It's been an interesting exercise and discussion for me though - all part of my continuing education. Right now I'm off to try out my new GND filters

Dave

[xpatUSA]

Unfortunately it seems like no other fellow forum use has noticed this mistake. Maybe this don't means too much to you but as an enthusiast of mathematics and formal logic I felt compelled to point this issue.
Regards and good luck.

I did - but had already tried to explain in another thread:

I find histograms more understandable by thinking about they are, rather than what they do in response to all those different kinds of pictures. The analogy to that thought is when someone plays different guitar chords to you it teaches you very little about about music theory per se. So, working from histogram to image:

The horizontal axis represents brightness and is 256 screen pixels wide. Each screen pixel is at the bottom of a column that represents a quantity, i.e. total number of occurrences, of image pixels ofexactly the same brightness, starting with 0 (black) on the left and ending at 255 (white) on the right. The histogram is actually showing how all of the image pixel values of 0 to 255 are distributed in your picture. The vertical axis is a bit harder to grasp because it has to cope with images having millions of pixels in them, down to thumbnail images with only a few hundred. For this reason, the vertical axis shows relative quantities and is self-scaling. It works if you think of it as a "normalized" graph where the area of the graph, i.e. distribution, represents 100% i.e. all of the pixels in your image.