Does your format include the sync pulses? For composite video you can look for correlation between the hsync and the rest of the line, forwards and backwards. If the ghost delays are bigger than one line, you need to use the vsync too! With bad ghosting on TV, you often see a darker vertical stripe on the left hand side of the screen. This is the ghost of hsync past! cheers, Syms. geschemann@yahoo.fr (Guy Eschemann) wrote in message news:<b9f16a5b.0310080818.74a45ed9@posting.google.com>...> Hi, > I'd like some advice on how to determine if a video signal has some > echos (ghost pictures) in it or not. > What I have is a digital video stream (ITU-656), and I'm looking for > some kind of DSP algorithm (line-based or frame-based) to solve this. > Any suggestions ? > > Many thanks, > Guy.
How to detect echos in a video signal ?
Started by ●October 8, 2003
Reply by ●October 15, 20032003-10-15
Reply by ●October 15, 20032003-10-15
OK, thanks to everyone who answered my original post. There a two things I forgot to mention : - I don't need to cancel the echo(s), just knowing that the picture has one or more echos in it would be enough for now. - I don't want to rely on a training signal provided by the transmitter. All the processing has to be done on the digital video stream. To better illustrate my problem I put a sample picture affected by such an echo at the following address : http://www.guy-eschemann.privat.t-online.de/echo.jpg I already tried to compute the autocorrelation function for each line (720 samples) of the picture, but wasn't able to detect anything with it. Someone suggested the "mutual information" method. Any thoughts ? Regards, Guy.
Reply by ●October 15, 20032003-10-15
Guy Eschemann wrote:> OK, thanks to everyone who answered my original post. There a two > things I forgot to mention : > > - I don't need to cancel the echo(s), just knowing that the picture > has one or more echos in it would be enough for now. > > - I don't want to rely on a training signal provided by the > transmitter. All the processing has to be done on the digital video > stream. > > To better illustrate my problem I put a sample picture affected by > such an echo at the following address : > http://www.guy-eschemann.privat.t-online.de/echo.jpg > > I already tried to compute the autocorrelation function for each line > (720 samples) of the picture, but wasn't able to detect anything with > it. > > Someone suggested the "mutual information" method. Any thoughts ? > > Regards, > Guy.The usual correlation probably won't show much, as you found. The ghost that's readily apparent to the eye is a correlation of waveforms at different amplitudes, something that pixel-for-pixel correlation won't reveal. This is essentially a pattern-matching problem. It may be easier than most because size and orientation are known a priori. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●October 16, 20032003-10-16
geschemann@yahoo.fr (Guy Eschemann) wrote in message news:<b9f16a5b.0310151232.433296db@posting.google.com>...> OK, thanks to everyone who answered my original post. There a two > things I forgot to mention : > > - I don't need to cancel the echo(s), just knowing that the picture > has one or more echos in it would be enough for now. > > - I don't want to rely on a training signal provided by the > transmitter. All the processing has to be done on the digital video > stream. > > To better illustrate my problem I put a sample picture affected by > such an echo at the following address : > http://www.guy-eschemann.privat.t-online.de/echo.jpg > > I already tried to compute the autocorrelation function for each line > (720 samples) of the picture, but wasn't able to detect anything with > it. > > Someone suggested the "mutual information" method. Any thoughts ? > > Regards, > Guy.Why not try cepstrum analysis. That method can pretty easily deal with discrete echoes. In the example picture I would expect to see a pretty large peak in the cepstral domain corresponding to the major delay. For large delay echoes you could probably get a reliable indication of ghosting. For shorter delays you'll most likely run into cepstral overlap issues whereby you can't readily say whether a peak is due to ghosting or is just an intrinsic part of the unadulterated image. I guess you could get away with a 1d analysis but 2d cepstrum analysis would probably give better reliablity. Regards, Paavo Jumppanen, Author of AtSpec : A 2 channel PC based FFT spectrum analyzer http://www.taquis.com
Reply by ●October 20, 20032003-10-20
Jerry Avins <jya@ieee.org> wrote in message news:<bm1tfn$n4j$1@bob.news.rcn.net>...> Eyes are pretty hard to beat with a program. I had a nasty negative TV > ghost from a nearby water tower. I fixed it by hanging an open stub in > parallel with the downlead, and trimming its length to move the positive > ghost it makes onto it. Then the right resistor to terminate the stub > pretty well cancels it altogether. >Jerry, as I am an electronics neophyte, I am interested in more details about this TV ghost "fix" (this is the sort of thing that can make one a legend among friends and relatives)! 1. open stub = a wire? What does it connect to? 2. what is a downlead? (assume we are talking about TV 'rabbit ears'?) 3. where does this terminating resistor attach and what's a rough initial value? Thanks!
Reply by ●October 20, 20032003-10-20
cb wrote:> Jerry Avins <jya@ieee.org> wrote in message news:<bm1tfn$n4j$1@bob.news.rcn.net>... > > >>Eyes are pretty hard to beat with a program. I had a nasty negative TV >>ghost from a nearby water tower. I fixed it by hanging an open stub in >>parallel with the downlead, and trimming its length to move the positive >>ghost it makes onto it. Then the right resistor to terminate the stub >>pretty well cancels it altogether. >> > > > Jerry, as I am an electronics neophyte, I am interested in more > details about this TV ghost "fix" (this is the sort of thing that can > make one a legend among friends and relatives)! > 1. open stub = a wire? What does it connect to? > 2. what is a downlead? (assume we are talking about TV 'rabbit ears'?) > 3. where does this terminating resistor attach and what's a rough > initial value? > > Thanks!I'm talking about an antenna on the roof. The downlead is the wire from the antenna to set, either 300-ohm twin lead, or in my case, coaxial cable. A stub is a side branch. It can be connected anywhere, but one end or the other os usually most convenient. An advantage of coax is that it can be rolled up. Signals bounce of mismatched terminations (cable ends.) If the end is open, the voltage bounces back in phase with the arrival and unattenuated. If the end is shorted, the return is still unattenuated, but the phase is flipped 180 degrees. When there's a resistor at the end of the stub that matches its characteristic impedance -- 300 ohms for twin lead, 75 ohms for coax -- there's no reflection at all. So you can adjust the timing and amplitude of the return signal by setting the right length and terminating impedance so it just matches the negative of the ghost, canceling it. A mismatched lead-in can create a ghost by itself; that's the easiest to fix. The length of the stub should be half the time of the ghost. (Apples and oranges, but lets convert.) The signal travels at about 80% of its free-space speed on coax; faster on twinlead. One scan line is about 63.5 microseconds. The free-space speed is about 1 nanosecond per foot. A ghost that's a quarter way down the screen needs about 8 microseconds one-way delay down the stub, or more cable than you probably have lying around. But it can be done. (Terminating with reactive components van make the the stub appear to be longer than it really is.) Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
