[mythtv-users] mythcommflag is pretty darn amazing

[David Schmidt]

On Tue, Sep 1, 2009 at 1:41 PM, Paul<pturpin+mythtv at gmail.com> wrote:
> On Tue, Sep 1, 2009 at 2:27 AM, Yeechang Lee <ylee at pobox.com> wrote:
>>
>> MythTV's commflagging has worked quite well for me for almost four
>> years, with very few false positives or false negatives. (I use All
>> Methods and Strict on -fixes.) But it managed to impress me again
>> today.
>>
>> I began recording from CNBC HD a few days ago. For those who haven't
>> seen it, various scrolling banners, charts, and other data occupy one
>> third of the screen at all times. Nonetheless, mythcommflag accurately
>> picks up the commercial breaks. How in the world does it do this? I
>> know it can't be the logo detection; is it the blank frames?
>
> From some work i did on the subject, here in the US at least, Blank frames
> are indeed the best indicator of what is and what is not a commercial.
> Actually it would be more correct to say the distance between the blank
> frames is the best indicator. Logo detection and such just helps to bump up
> the accuracy a little bit.
>

Anyone who regularly records "Jeapordy" in the US definitely knows
it's mostly blank frames that drive commflagging.

At least here in Dallas, they always have a row of commercials from
syndication sponsors just before the final question.  To my eye, these
are exactly like the regular commercials--no channel logo, rapid cut,
etc.  Only difference is a blank stretch between the local commercials
and these and none between these and the question.  Myth ALWAYS
includes these as part of the actual "program."