A New Approach to SYSM=3 Image Processing
[Minor revision. See description here
Several of us have expressed frustration while trying to optimize Sharpness settings with SYSM set to 3. With SYSM=2, it has seemed easy. But SYSM=3, while appearing smoother, also gives rise to coarse oversharpening or overshoot while not handling fine textures properly. Can it be improved?
Indulge me here in some audio graphic-equalizer and frequency analogies. Think of fine detail as high frequencies, higher => finer detail. Overshoot or ringing that's fairly coarse can be simulated by raising the mid-treble without raising anything else, boosting a *narrow range* of high frequencies in a hump, but not the highest frequencies a bump in the frequency response, as it were.
SYSM=2 appears to configure the Sharpness slider so that it boosts high frequencies in an upward-sloping ramp: the higher the frequency (finer the detail), the more boost. This looks natural on-screen when tuned just right and tends to counteract natural losses in the video chain, since they tend to be manifest as a ramp-downward in the high frequencies. So far, so good.
But what is so different about SYSM=3? Looking at it intuitively, it appears to me like SYSM=3 configures the Sharpness slider to boost detail starting in, say, the mid-treble (think audio), but does so as a *plateau*: extreme highs (finest textures) are boosted along with mid-highs (mid-detail), but no more than mid-highs like a ramped-step up instead of a sloping ramp that goes up forever. Doesn't look natural.
What might improve this? Enter the interesting high-frequency filters in MID5, MHLY
. When you first put up a resolution test pattern and vary MHLY from 0 (off) thru 3 (max), it seems just to attenuate detail, like turning down the treble control except for MHLY=2. Think of that graphic equalizer again. Beginning at, say, 2000 the frequency response begins to slope downward, but it *levels out* at 8000 instead of continuing to slope downward forever. MHLY=2 does not seem to be the same-shaped filter as 1 and 3: it's much milder and seems not to trash the finest textures. (MHLC appears to do the same for color.) Well, well . . . Does this sound like the opposite of SYSM=3 or what? What if they complement each other somehow?
(Yes this is how my mind works! Blame it on an early interest in Science, niggling detail, and exposure to all those lab chemicals and lead solder.)
Here are some suggestions for 2170P-3 and MID5 settings that have resulted in less ringing and overshoot than anything I have yet found, and yet super-fine detail is preserved! Use my chart posted here
to make sense of the following settings listings. I am listing only setting-columns for 2170P-3 and MID5, and your VM setting doesn't matter here. Typically I have Sharpness set to 30-ish. The "0-3-0-8" group are my favorite VM-shaping settings these days.Typical HD Broadcast @ 1080i for SYSM=2.
2170P-3: 2n0-3-0-80-1-3-00-1-0x-y-z61 where n,x,y,z = original settings peculiar to my TV.
I then set up a different picture mode identical to the above, with a few differences:
where 40 is a new MID5 column for this purpose.
-00-0-00-0-0 (Also try 0-3-4-0 for the 2nd and 3rd groups.)
Tested this on the winter Olympics (indoor venues) with stunning results. Photo-like fine details without oversharpening! Gorgeous! I switched among this, the usual SYSM=2 settings, and the more enhanced 0-3-4-0 also suggested above. On slightly softer HD material, I wouldn't hesitate to recommend 0-3-4-0 in MID5 for the 0-3-3-0 groups. Further watching over the last two weeks has confirmed that these are really good settings, too.
I also tried this technique on ordinary SD material, tested with impeccably broadcast evening news on local stations.Typical SD 480i settings for SYSM=2:
For the test mode I then set up these values:
I also set up a stronger sharpness enhancement with 0-3-3-1 replacing the 0-3-2-1 in the MID5 settings. (Yes, I have set up two of the picture modes (Pro and Vivid) strictly for testing. I use Standard for SYSM=2 settings and Movie for lower Gamma. But that's another discussion. None of Sony's original picture-mode settings have been retained.)
This has transformed my appreciation of SD programming. When the broadcast is good, there is less grunge and greater clarity with fine, photo-like detail (as good as can be had from SD), especially with the 2103-1 tuning outlined in a recent previous article. When switching back to my SYSM=2 mode, I see a different kind of sharpness, just a bit more scratchy, a bit less attractive, a bit less three-dimensional. And with a bit more overshoot.
There is a systematic method
to these new settings you can apply to whatever you already have set up for yourself. Here are the rules I'm following, applied to both 2170P-3 and MID5:
I'm taking my favorite settings for any one particular video mode, say DVD over V5 or HD-720p, you name it; and I am:
(1) Changing SYSM=2 to SYSM=3.
(2) Raising 2170P-3/SHOF from 0 to 3 to compensate for the Sharpness slider differences required.
(3) In MID5, changing both MHLY and MHLC from 0 to 2.
(4) In MID5, adding one or two ticks to the sharpness parameters MHYE and MHCE. In the case of SD material, I am adding a *tiny* bit of vertical sharpening by changing MVYL from 0 to 3 while leaving MYVE at 0.
I'd be grateful for any feedback.
(1) Many thanks to Nitewatchman
for PM'ing back and forth with me to help me understand I wasn't hallucinating these improvements. They may be subtle, but they're not imaginary! He also helped check out the method for tuning the 2103 settings in the previous article.
(2) I have now taken over MID5 columns 30-63. I see no more need to preserve Sony's original settings, as they have been superseded in every instance by my own. For that matter, 0-29 aren't sacred, either. If any column ever has to be restored, the SM charts have all of the default values.
(3) In the above examples, the MIDE/MID5 values can be anything, as long as they agree. Use whatever column(s) you have already configured, or a new one, it doesn't matter. For example, in my scheme MID5 column 34 is now permanently set as indicated, and that's what I show here as an example. But you could use any column, not necessarily 34. Same with 55, 40, etc.
(4) These experiments are not casting any settings in stone but pointing in a certain direction, and I sure like the results. Some further refinements may be necessary. I'm hoping there are folks out there who have tried out some of the settings we've suggested before, working with the IP chart, who will try these, too.