Ad Blocker

I haven't seen the letter you mention but, as you say, doing it in hardware sounds difficult. How are you planning to do it? I guess that you'd need to capture and store the last bit of the last few lines in a frame then use digital image processing techniques to detect the logo, using a flash convertor and memory and a CPU.

Have you considered doing it in software with existing hardware? Perhaps a PC with a TV card and software that periodically checks the captured image for the logo, then switches channel or interfaces to the remote control or whatever.

Eitherway, a neural-network detector might be fun to try.

Cheers,

Cheers,

I looked into this about 20 years ago. Back then, most of the TV stations fade to black for a brief time, before and after ads. Perhaps this may be a simpler method.

Trevor Wilson

sounds difficult. How

last bit of the last few

logo, using a flash

Yes, that is basically the approach I had in mind, except for the image processing bit - I am hoping to digitise a line or lines and look for a value that changes as the logo appears/disappears

with a TV card and

switches channel or

I had a look at doing it in software but you really have to get 'down & dirty' to process the image & Windows puts up massive barriers between you & the hardware. You could, of course look at doing it in DOS, but then you can't get drivers for available capture hardware, and I don't have the development tools. I do like the idea of a neural network though - I will have to do some more research.

Dave

Dave Goldfinch wrote in news: snipped-for-privacy@4ax.com:

Why on earth would you lower yourself to doing it with *any* microsoft or microsoft-derived product when there are any number of unix and unix-like operating systems available completely free of charge that will give you direct access to all you need to develop working code, and numerous libraries that will do much of the work for you.

GB, I have problems comprehending the ongoing use of microsoft products in business, but this, this is just... speechless!

I guess you have never worked for a commercially succesfull software business.

I think that the logos are a sort of watermark thing; the logo pixels are reduced in intesity compared to the underlying image pixel. They're not of a fixed intensity. This could make detection pretty tricky, especially as the underlying image could be changing too.

Edge detection followed by logo-specific detection might be the go.

As someone else has pointed out, using an Open Source unix, most likely linux, is a much better option for this sort of project IMO, especially if it's dedicated hardware. I can elaborate if you'd like.

I reckon that a reasonably small, bog-standard back-propagation neural network would do the job. I'd use a sampling of the edge detection data for the inputs. You'd probably have to train a net for each logo, but then again maybe not, then again the net might take care of the edge detection too.

Cheers,

One thing that might put the brakes on your hardware idea - at least one station slowly moves their watermark within a small area in order to avoid plasma burnin. You'll probably need more processing power to figure it out.

there isn't always _one_, I think you'll need to do edge detection and if you see enough edges in the right locations you're seeing the logo,

locations will vary form station to station, many logos are transparent..

with a TV card and

switches channel or

that's what drivers are for.

many dos tools are free.

figure on 4 or more neurons per pixel, I don't know where to get nneural network hardware.

Bye. Jasen

the logo for each station is in the same place? so you really only need to examine a small number of pixels?

Well, how do you add a watermark? Is it just a signal mix Can you detect that signal in the logo area?

would it be a hue/saturation/?? difference

or is it something simple as 50% of pixels become a set colour over the area. that should be easy to detect

there was a package called motion (since renamed) that was designed to detect movement in a cam and start recording. It did this by % change in scence. I was thinking it might just give you the shell of the program and you just change the analysis to report what happens in the logo area.

Just some idea. Not my area.

Elektor had a design in 2004 that worked on this basis:

It used an embedded micro with a learning IR tranceiver to transmit a 'pause' signal to the VCR.

Would require modification to use in Aus because it looked for the logo in the top 1/4 of the picture (as in Europe, aparrently).

FWIW.

Chris.

Hi Chris

Thanks for that - it appears to do exactly what I would like.

At least this suggests that it is possible to use the logo, not just a hopeless dream.

Dave

If you digitised the image then there are some math functions that could look at that corner of the image and pick up a sudden fixed change in the value from one pixel to the neighbouring one caused by going from a non watermark area to a digital watermark.

The key would be to look for a relatively fixed number of pixel transitions changing by a fixed amount up (whiter) and almost the same number of transitions going down by the same fixed value for every scan line.

You would have to play with tolerances a bit but it should work. The hardware would be complex though, possibly a PC and frame grabber.

Dave Goldfinch wrote in news: snipped-for-privacy@4ax.com:

Forgive me to leaping into this thread so late, but as a matter of interest, Phillips recently gained a patent (In the US I suspect) for a system that would force punters to watch advertisements by disabling pause/ffwd/channel-change/etc functions for the duration of the advertising.

Notwithstanding that the company made noises about their intent being to lock it up to protect anyone from actually doing it, there were there to-be-expected round of "patent on stopping people from buying Phillips televisions, etc, etc" whinges.

All the Free Software/Open Source Software folks who are involved in authoring PVR type code took a very different view. Theirs was "Thanks Phillips, for creating a system that tells us reliably when the advertisements start and end"!!!

GB

I have the original article if you're interested.

Dunno if I have the CD (with article in PDF format) for that year - I'll have a look. Otherwise I could probably scan it for you - if you have no deadlines to meet (read: it may take some time!) ;-)

Still for the price of ~1 quid from their website...

Chris.

Thanks for the offer. I went & had alook at their website & was about to download the article but was put off by their paying arrangenments

- they seemed to want far too much personal detail - I am probably just a bit paranoid about disclosing card details over the net, but this really seemed over the top. Have you used this service ?

Fortunately the WA library service gets Elektor & so I have just been into town & got a copy of the article - will sit down now to read & digest.

Dave

Since I've been getting the mag for a few years now, I haven't had to, no. It is annoying when they want all those details, & there's always

Ah, a library... that's that big sort-of shop thingy where they let you borrow stuff - for nothing! - innit? ;-D

Happy reading!

Chris.

ads

logo

Yes, but you would miss the start of each segment as the logos are often slow at coming back. Surely the FF control is a lot cheaper anyway.

MrT.

all that's needed for a vertical edge detector is a differentiator (high-pass filter) and a pair of limit detectors. (one for light-to dark, one for dark-to-light)

then you just need to spot edges that are in the right place.

if you get 90% or so you've probably got the logo.

Apart from the possible inconvenience of waiting for a new card, what is the risk? If someone uses your credit card fraudulently the issuer will bear the cost. In any case, there is probably far less risk of that than someone obtaining the details from the hard copy you used in a local restaurant.