12 - OPTIMIZING IMAGE DECODING, "SHAPING," AND ENHANCEMENT
This has been the most difficult article to write, since it involves *subjective* judgments of image quality. So, up-front, my goals here were to try to get the images on my 36XS955 to *not* look like video but like realistic, high-quality film, at its best, where possible. Maybe even reality! The hardest part to get right is the reproduction of *texture*: human skin, clothing, rock and sand, distant trees, hair. Many image enhancements smear or attenuate this texture, but I am happy to claim that these particular Sony sets seem capable of reproducing it very well - providing it's not killed within the set by misguided filtering and over-the-top enhancements. What have helped me are (1) an extensive photography background, and (2) expectations in line with reality. TV is a crummy picture by most standards, but there's no reason one can't maximize its quality for what it is. And HD raises the stakes.
Video can have a look, and I have typically found it unattractive - overly bright, compressed luminance range; over-sharpening of all detail (historically called aperture correction; exaggerated color, especially food colors and reds; and *way* over-sharpening of high-contrast edges independent of other detail. Speaking optimistically, I have found that, even among SD channels, there is a great deal of first-rate NTSC video, whether local news, Survivor on CBS, or some documentary on the History or Food channel. I don't want the set tinkering excessively with good video. Just pass it to the screen, thanks, with suitable noise suppression and a little sharpness enhancement to compensate for losses in the long signal chain from camera thru editing to broadcast.
So I am striving for *transparency.* Some broadcast and VCR video is hopelessly bad, and so it will appear on-screen. It seems that attempts at manipulation just make it worse! So transparency works for me there, too, and I just deal with it.
There are too many technical details to write about in a forum message, so I have attached a PDF document converted from a Word document of the real article in its entirety. A detailed chart expresses succinctly what I have discovered about the very important 2170P-3 settings and the related MID5 chart, and I note a couple of newly-discovered (9-8-05) settings in 2103-1. (The charts' boundaries may print better than Adobe Reader will display.) In addition, please download one of the service-data charts available in this thread before attempting to make changes, as it makes plain the relationships among the picture and video modes and these settings. (Note: No discussion here about gamma, black level, color, geometry, convergence, etc. See other articles.)
That said, this continues to be a fascinating discovery process, and experiments can be conducted *safely* by (1) using one of the useless picture modes for playing around and (2) by laying claim to a couple of the columns in the MID5 table for your own purposes. If nothing else, I think just trying out the settings suggested in the table will really please some folks who are searching for this image perfection.
Feedback from readers is VERY WELCOME. Please! And let me know if there's any trouble with the PDF document.
[Small changes to message and major revisions to PDF document 9-8-05.]
Image_Processing_rev1_1.pdf 147.017578125k . file