|
|
Posted by Toby on 07/13/06 01:30
"PTravel" <ptravel@ruyitang.com> wrote in message
news:4hfr7qF1rrhbgU1@individual.net...
>
> "Martin Heffels" <youwishyouwouldknow@nottellinya.com> wrote in message
> news:o7b5b25qf7up8n6ffg8sj9oj76ehur1ovn@4ax.com...
>> On Mon, 10 Jul 2006 17:27:07 GMT, "PTravel" <ptravel@travelersvideo.com>
>> wrote:
>>
>>>But that's the whole point -- if it's corrected, it's not a drop out, and
>>>there's no data loss.
>>
>> No, it is not. Because if the drop-out is severe enough, data will be
>> copied from another part (interpolated from adjacent frames or areas),
>> and
>> that means you have dataloss when compared to the original.
>>
>> From Keith Jack's "Video Demystified":
>>
>> <quote>
>> Error Concealment
>>
>> The ability to hide transmission errors that corrupt the content beyond
>> the
>> ability of the receiver to properly display it. Techniques for video
>> include replacing the corrupt region with either earlier video data,
>> interpolated video data from previous and next frames, or interpolated
>> data
>> from neighboring areas within the current frame. Decoded MPEG video may
>> also be processed using deblocking filters to reduce blocking artifacts.
>> Techniques for audio include replacing the corrupt region with
>> interpolated
>> audio data.
>> </quote>
>
> That's still apples and oranges. I didn't say that this kind of
> "correction" isn't done. I said that it's extremely rare and the kind of
> common "drop out" that you've been discussing is fixed by ECC in such a
> way that no data is lost.
You guys are talking at cross purposes. Martin's point is that there are two
types of error correction, one in which ECC reconstructs bit-perfect data
based on redundancy, etc, and another where it corrects data in a
possibly-undetectable, but non-perfect way, such as pixel interpolation,
etc.
I'm not saying it happens "often", and I haven't found anything useful on
the subject yet, but it seems entirely plausible that errors of the second
type could be cumulative over a number of copies (whatever their cause,
whether because of corrupted physical data or read errors or quantum errors
or cosmic rays hitting circuits or whatever), and they could eventually
become detectable after some threshold has been crossed.
For example, no one is able to discern audio differences of 1 dB, but do
that six times so that the difference is 6 dB and almost everyone would be
able to detect that.
As long as some sort of lossy ECC exists, after some number of passes it
seems logical that it will build up to a point in which image degradation
will be detectable even to the most stubborn lawyer ;-)
Although I'm not sure that 18 passes is really the threshold.
Toby
Navigation:
[Reply to this message]
|