Histogram Transform: graph from top does not move down after applying

[Nocturnal]

Hi,

when you're fiddling with the HT process the top graph shows what the histogram will be like after you apply the current transform, right? If that's true then why does the top histogram graph not move to the bottom graph after I apply?

[Juan Conejero]

Because your first statement, namely:

the top graph shows what the histogram will be like after you apply the current transform

is true. So after applying HT the top graph predicts the histogram as it would be if you'd apply the same HT again. The bottom graph always shows the current histogram functions.

That said, always bear in mind that the top graph is a prediction generated from histogram data, obviously not from pixel data. There can be some small differences with the actual histogram after HT, which usually are negligible.

[Nocturnal]

I think you'll agree there is a *huge* difference between the top graph in the first image and the bottom graph in the second. Why is that so? Is the estimate in the first that far off?

To be clear, the difference between the two screen shots is: apply HT and reset HT parameters even though the reset of course does not affect the bottom graph at all. When I apply the HT I would expect the top graph to move to the bottom.

[Juan Conejero]

Sorry, I didn't look carefully at the screen shots in your first post. Now I understand what you mean.

The prediction in your example is correct. It just happens that RGB histograms are always shown normalized. Normalization means that the R, G and B peaks are drawn with heights proportional to their relative amplitudes. The highest peak occupies the whole graph (vertically), while the other two peaks are scaled proportionally. In this way you get visual information on relative peak intensities. However, sometimes the final scaling ratios cannot be predicted accurately by HT's top graph, and hence you can see differences as in your example. This is quite unusual though. Note that the positions of the red, green and blue peaks after HT match the predicted positions very accurately. So this is not a bug, but a limitation of the histogram prediction routine.

[Nocturnal]

Ok, thanks Juan for taking another look and explaining it to me. I like to tell people that the HT process alone is worth getting PI for so I like to understand what it's doing

[Juan Conejero]

The whole design and a 90% of the functionality of the HT tool are very old; I think they haven't changed since 2003. Note that this is even older than the first PixInsight LE release (2004)

I appreciate your support. For many years I've been unable to understand how and why so many users of other well-known applications can live with so inferior (embarrassingly poor I would say) implementations of a fundamental tool such as the histogram transformation. And even more when we are speaking of a discipline where the histograms are so complex and require so accurate and drastic manipulations. Isn't it curious?

[Nocturnal]

I think the problem is many people don't truly understand what the HT does and thus don't get why you want 64K samples and both pixel and percentage counts to show how many pixels you're dropping.

[RBA]

I think the problem is many people don't truly understand what the HT does...

No, the problem is a lot more about human nature than some technical details.

[RobF2]

Yes, most people seem to get excited about DBE in PI before anything else.  I often get a funny look when I say how fantastic the HT tool is.  It's like "what's the big deal - how hard is it to do some stretches in PS?".  I suspect experienced PS users intuitively get a feel for where to put the stretch points, but it's so much smoother and reproducible in PI.  Interesting to hear that functionality dates back to 2003.  Sounds like Juan - get the fundamentals right and keep add more and more blocks on top as you reach for the sky.... 8)

[martin_magnan]

Hi guys

I have a question about this subject.

If the image we apply the HT is masked, does the calculated histogram take the mask into account or not ? I suspect it does not.

Martin

[RBA]

If the image we apply the HT is masked, does the calculated histogram take the mask into account or not ? I suspect it does not.

No, it doesn't.

[Nocturnal]

No, it doesn't.

I think it should

[Cleon_Wells]

While we’re kicking the HT tires.
Why does increasing the HT vertical graph gain show me the top of the display,  so I then have to pull the vertical adjustment slider down to see the bottom of the display?   (  I think I’m getting that dreaded Carpal Tunnel Syndrome, 8>)--
Increasing of the horizontal gain show what I want to see, the black point, without moving the horizontal slider.
Cleon

[Juan Conejero]

If the image we apply the HT is masked, does the calculated histogram take the mask into account or not ?

No it doesn't. The top histogram graph tells you how the adjustments you're making to HT parameters would change the current histogram, if applied to the image directly without any mask. For more sophisticated evaluations, you can create a preview on the image and try HT on it, then use the undo/redo preview command to compare the before/after histogram on HT's input histogram graph.

I think it should

Note that in order to generate a real-time prediction of the output histograms (that is, automatically updated as the user changes HT parameters) after masking the processed image, the whole process, including masking, would have to be applied to a temporary duplicate of the image. The HT tool could generate reduced temporary duplicates on the fly to implement this functionality, but this would be just a rough approximation. This is one of the tasks for which previews have been conceived and implemented in PixInsight.

[Juan Conejero]

Why does increasing the HT vertical graph gain show me the top of the display

Yes, this is something that should be improved. The zoom function should preserve the current position on the graph, and if zooming in from 1:1, the graph should stay anchored at the bottom left corner. I'll try to fix this issue with an update to the HT tool. Thank you for (re)pointing this out.

However, if instead of zooming in/out with the zoom spin boxes you use the mouse wheel, then the graphs try to preserve the position at which the cursor is when you move the wheel. With the graph at 1:1, place the mouse near the bottom left corner and move the wheel towards you. This will zoom in and the graph's position won't change.