[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: DVNR



Craig,

I would like to offer some info on the DVNR statements made by Marc.

First let me preface this with a disclaimer:  My company provides installation
and service for DVNR products (as well as POGLE and Vistek) in the US, but I am
NOT a salesman. I know a great deal about these products, and have worked with
them for several years now. I will be straight with you, as my background is
from managing level of engineering for several of LA post houses for many years.

That out of the way, here goes:

The "artifacts" to which Marc is referring on titles are caused by the Scratch
and Dirt Concealer, not the Noise Reduction. This option set of boards works by
detecting "impulse noise" on an individual frame, then replacing it with a
median value from the same pixel locations on the two "bookending" frames (i.e.
the frame ahead and the frame behind).

Since the analysis circuitry is not perfectly intelligent, fine detail that is
moving can be mistaken for "dirt", and replaced accidentally. This will cause
the "deletion" effects of which Marc spoke.

Allow me to give a brief description of how this works. The detector circuit
compares three frames at a time:  current frame, one behind, one ahead. If it
finds a small group of pixels, the size of which is determined by the operator
from four choices, that is near black or white (dirt/scratch) for only one
frame, it assumes that this "impulse noise" is a dirt bug or a small scratch.
The operator also has the ability to affect two other parameters: contrast and
motion. The contrast sets the "difference value" between the dirt and the
surrounding video. A low setting insists on a high contrast between the dirt and
surround, as the contrast value is increased, a lower difference value is
allowed. You would need this, for example, to detect white dirt against a light
blue sky. The other control sets the amount of attention the detector pays to
moving objects in the picture. Since a moving edge can, under some
circumstances, be mistaken for dirt appearing and disappearing the detector can
be made to relatively ignore the moving parts of the image. However dirt is very
democratic, and shows no predilection for only the still parts of the image, so
this can be altered.

Since the "serifs" (little edges of letters) can be only a line or two high by a
few pixels in length, it is very possible when these are moving, for the
detector to think that this is dirt coming in and out. The replacement portion
of the circuit will then oblige and fill in this area with a median value of
frame before and after. Since this is often black surround on white titles, bits
of the letter are effectively erased.

However, Marc is completely correct in that the overall _resolution_ of the
picture is unaffected. The only errors are in the small pixel blocks that are
replaced. This is in contrast to some other devices that use median technology
on the whole frame, causing actual resolution loss to the entire picture.

There is no perfect box of filters. Anybody that says there is, is probably a
salesman (grin). All such filters have a range of effectiveness that must be
balanced against the artifacts produced. I will be the first to tell you that
every one of the filter options on a DVNR, will under the right circumstances,
produce objectional artifacts. But so does every other filtering device. The
correct application of a judicious amount of filtering can make the end product
better, but careful tending by a skilled and knowledgeable operator is a must.

I hope that I have not been too long winded, but I feel that an accurate
understanding of the technology involved can quickly put to rest a lot of rumor
and guessing. I am open to comments and questions.

Best regards,

Ed Elliott