George,
This latest image looks very good! And I must say, your DofFtest.html file may be the prettiest problem report that I have ever seen. It's nice to watch an artist at work.
Your problem with CombineZ5 is probably fixable by adjusting a noise threshold. If you Macro | Edit Macros and select Do Stack, you'll get a Macro Editor dialog that shows about 17 individual steps. Around step 4 is "Find Detail(25)". (These numbers may change from release to release.) Highlight that line and push the Edit button. Near the bottom of the dialog, in a box labeled Parameters, change the "25" to a "5" or some such. Then push the Update button, and the line in the list will change to match. Push the OK button to dismiss the dialog. Then re-run your stack using Do Stack, and see what happens.
Some explanation will help understand what's going on here. Helicon Focus and CombineZ5 use very different algorithms.
After registering the images, both of them evaluate around every pixel in every frame to get some "sharpness" number that reflects how crisply that part of that frame appears to be focused.
After that, the algorithms are completely different.
Helicon Focus (HF) proceeds to weight every frame by some nonlinear function of sharpness, independent of stacking order. What appears in the composite result at each pixel position is a combination of pixel values that heavily weights the sharpest frame(s), regardless of how fuzzy that frame (or frames) might be.
CombineZ5 (CZ5), on the other hand, decides at each pixel position which single frame is the sharpest and assigns that frame number as that pixel's "depth". However, if all frames are below the Find Detail threshold of sharpness at a pixel position, then that pixel position initially gets assigned a depth of "can't tell". CZ5 then fills in all of the "can't tell" positions by interpolating the smoothest surface it can that still fits all of the depths that could be determined. The final filled-in depth map is used to select the source frame at each pixel position, interpolating between pixel values for depths that are not integer frame numbers.
In your case, I believe what's happening is that the entire left background is sufficiently out of focus that it's falling below the Find Detail threshold. The initial depth map has that whole area as "can't tell", and the interpolation process fills in from the nearest pixels whose depth can be determined, many of which are from the point of the screwdriver bit. By reducing the Find Detail threshold, you allow CombineZ5 to determine the depth even in that somewhat fuzzy region.
Understanding the different algorithms often helps to clear up why HF or CZ5 works better in specific circumstances. The big win of the CZ5 algorithm is that it has a lot less problem with haloing against fuzzy backgrounds, if you tune the Find Detail threshold so that halos fall below the threshold. But the CZ5 depth map algorithm fails badly when the subject does not have an obvious surface, say because it's a loose mat of fibers or a bunch of individual bacteria distributed as a 3D cloud. In those cases, CZ5 tends to produce a characteristic "swirly" effect as the single-valued depth map does its best to slew from one depth to another. Wrong model, wrong result.
By the way, in the interest of full disclosure, I should tell you how I know this stuff. With CombineZ5, I had the great fortune to collaborate with its author Alan Hadley in developing the depth map algorithm. For Helicon Focus, all I know is what I can tell from experimentation and comparison with my own earlier work. If you get results that suggest something different is going on, please let me know. Both authors have used some of my stacks as tests and drivers for their software development.
Hope this is helpful,
--Rik
