Transcript: Video Tutorial #2: Advanced Retouching

This is a transcript of the video tutorial shown HERE.

In this tutorial we will be looking at how to process this stack of honeysuckle blossoms. Because this stack was shot in the field and not in a studio, there are some interesting and challenging issues posed by subject movement, in addition to the usual issues that arise in all stacks.

The workflow that we will be using is this. First we will run the stack using the PMax method. PMax generates an output that gives us a good first look and is almost always helpful to have as one source for retouching. This first run of the stack will also align all of the frames and generate a bunch of cached images that will make further work go faster. Then we will study the input images for the stack to identify issues that we may need to deal with later. We will run DMap, perhaps more than once, and finally we will use retouching to put the best parts of everything together to make a final output.

What we are doing right now is looking at the original source frames. We can immediately see that there is quite a bit of subject movement, probably caused by just a little bit of wind. Honeysuckles are notorious for blowing around in the slightest breeze! If there were more detail visible in the background, this movement of the foreground subject could be a problem. But in this case with an extremely out-of-focus background and a high contrast subject, the alignment algorithm will probably concentrate on the subject and give us an acceptable result.

Let’s run PMax. This will take quite a while, so we will fast-forward to its completion.

OK, that didn’t take too long – about three minutes. Now, let’s see what we have.

First, let’s check the alignment. By press-and-drag within the list of input files, we can play through the stack almost as if it were a filmstrip or movie. For these first couple of passes, we’re looking at the input frames before alignment. As we noticed before, they’re moving around quite a bit.

Now, to check how well alignment worked, we can put a checkmark on “Show as adjusted” and repeat the process. The impression we get now is that only focus is changing. That’s good! At this point we really do not want to see any subject movement.

Unfortunately, we do see a little. Here at frames 3821 and 3822, and again at 24 and 25, there are small rotations of the subject that amount to changing the viewpoint. Those sorts of movements cannot be completely compensated by Zerene Stacker. What will happen, as we see here, is that even after alignment corresponding parts of the subject appear in slightly different places in adjacent frames. That is likely to cause problems in the stacked output, because any features that are equally in focus in both frames are likely to appear in two places at once in the final output. This is a type of defect sometimes called “motion echoes”.

Sometimes the subject will be out of position in just one or two frames. In that case, removing those frames from the stack may be a good solution. But the movement we see here doesn’t look big enough to justify that treatment. Let’s just forge ahead.

Looking now at the PMax output, we can see what looks like a good start, but as usual there are some aspects that can be improved. At this scale, the most obvious flaws are some broad dark halos that appear around bright features of the subject where it contrasts against dark background. Those are typical of PMax, but not DMap. Looking closer at PMax – all the way down to 100% scale (actual pixels), we see that there is quite a bit of pixel noise, much more than appeared in any of the source frames. Again, that is a common feature of PMax that is not shared by DMap.

At this point, it is almost time to run DMap. But first, there are some parameters to be set. The DMap algorithm is controlled by two “Radius” settings that must be set to match the sharpness of detail in the image. One good way to set those is to look at the PMax output at various zoom levels. What we see here at 100% is that even the best focused areas are just a bit blurry – the image is not really sharp at the level of individual pixels. On the other hand, if we scale out to 50%, then the image does look sharp. The default Radius settings are designed to work well with images that are sharp when viewed at 100%. Because these images are not quite there, we need to increase the Radius settings. A good rule of thumb is to make them bigger in proportion to the zoom factor needed to make the image look sharp. In this case, we needed to zoom out by a factor of two, so we will increase the Radius settings to be twice as big as the default values. We can do that by going to Options > Preferences > DMap settings, clicking the “Reset to default” button, then moving the sliders to twice their previous values. Click OK to accept those settings and dismiss the dialog.

Now let’s run DMap. While we’re at it, we can reset scale so that we’re looking at the full image as DMap runs.

As the DMap output forms up, we can see some halos forming around some of these high contrast features. Those are important things that we are going to get rid of by changing what’s called the contrast selection threshold, midway through the DMap process.

OK, here we are just half way through the DMap process. At this point, the algorithm has processed the entire stack over the entire image and is waiting for guidance from the human user about what matters. The underlying issue is that DMap only works well in regions that contain significant detail. In areas that do not contain significant detail, DMap will end up paying too much attention to noise and is liable to produce halos and blotches. What we need to do at this point is to adjust the slider so that we get clean borders, especially where the subject is silhouetted against very blurry background. In the right-hand panel, areas that are considered to not contain significant detail are shown in black. Our goal as users is to adjust the slider so as to make out-of-focus background go black, along with whatever halos might be in it, while leaving most of the subject shown in its original tones.

If we push the slider all the way left, we can see halos in out-of-focus regions. If we push the slider very far to the right, then lots of the subject goes black too, leaving mostly the borders. This setting may actually be useful in some areas, because it will give the cleanest borders around high contrast features like those yellow petals. But such a high setting is liable to miss detail in lower contrast portions of the petals. So let’s try for a happy medium and see where that gets us.

A setting around 40% looks pretty good in this case. That value will change from scene to scene, by the way, because it essentially says how much of the image is occupied by regions with little or no discernable detail.

When we click “OK”, we turn DMap loose to finish the rest of its processing. First there is a period of computation where we cannot see anything happening, and then DMap processes the entire stack again, starting with the first frame. During this process, we can see the result image form up.

OK. DMap is now finished. Let’s zoom in a little bit closer and see what we have. In the upper left corner, there is some visible banding in the out-of-focus background. This is probably due to changes in the pattern of illumination between input frames, typically caused by things like clouds going over that change the quality of the light, or shadows moving in the background. Around some of the petals, we can still see bits of colored halo. This is especially obvious in the central area, where we have little blobs of pink in areas that we know should be pure green leaves.

Most of those we will have to attack with retouching. But probably we can simplify our work by running DMap again with a different set of parameters that will give us cleaner edges. To do that, we’ll put the two Radius settings back to default, and set a high contrast threshold.

Here we are half-way through the second DMap, setting a contrast threshold that is so high it hugs tightly to the edges of those high contrast petals.

OK, the second DMap is now finished. At first glance it looks clean, but on closer study we can see that lots of detail has been lost in some areas. So, we have three output images, no one of them is ideal, but now we can start putting them together without too much trouble.

To begin retouching, we pick whichever output image has the most area that we like. In this case, a good place to start is with the first DMap. So let’s select that image as our base or “master” for retouching, and begin. In some areas, the PMax output image will be the best source for retouching. PMax also makes a good comparison against DMap, so let’s select that first.

Now, during retouching, there are several mouse and keyboard shortcuts that are very helpful to know. First is that whenever you are positioned over an image window, pressing the “s” key will flash between source and master. (“s” for source, release “s” for master…source…master… This is displaying in the right-hand window. source…master… It may be easier to see if we maximize that window. Source which is PMax, master which is DMap. Source is PMax, master is DMap. Let’s go back to side-by-side display.) Flashing makes it easy to identify areas that should be retouched. Second is that scrolling the mousewheel will adjust the size of the retouching brush. If you are using a mouse with no wheel, then the square bracket keys [ ] on the keyboard will do the same thing. Third is that whenever the source image is visible, pressing and holding the shift key turns the mouse into a depth navigation tool. With shift key held down, turning the mousewheel or press-and-dragging the mouse up and down will quickly select an individual input frame. Notice that it is critical to keep the shift key held down while you are doing this. Every time you release the shift key, several seconds are lost while the software prepares the currently selected frame for use as retouching input. And finally, pressing and holding the spacebar turns the mouse into a panning and zooming tool. In this mode, rolling the mousewheel changes the scale, and pressing and dragging the mouse pans the image.

With these tools at our disposal, it is convenient to do all our editing in a single maximized image window. However, for maximum clarity during this tutorial, we will continue to work in side-by-side mode.

OK, let’s get to the actual retouching. The pink blotches on the green leaves can be painted away using the PMax output as retouching source. So we select PMax, select an appropriate brush size, and simply paint over those pink blotches. (Pan to see more, paint over pink, from the PMax output. … Looks like there’s a little loss-of-detail halo here…more pink blotches there…bit of pink and a loss-of-detail halo there.) Banding in the background, and the halos around the outer petals, can be painted away from the second DMap output. So we’ll select that. This is the one made with default Radius values and a high contrast threshold. Make a large mouse…the second DMap output is sharp as well even a little bit into the interior of the petals, so we don’t need to worry about a little bit of overlap there.

Now I’m just looking over the images to see if there is anything else left to fix up. Usually there is something that is best done by going back to the original source images, but in this case I’m not seeing anything that looks like that. (Bit more pink blotch that needs to be fixed up from the PMax, right there…opportunity for a little bit sharper edge from the PMax…a bit of ugly blotching in there… Let me maximize now and look at the whole thing a little bit better…)

Looks pretty good, I think I’ll quit now.

Just Edit > Commit retouching, and we get a new output image. This would be a good point to save the project again, just in case we want to do some more retouching later, and of course at this point we can also save the final image as either JPEG or TIFF for printing, posting, or emailing.

And that’s the end of the tutorial. I hope this helps!