gwpoulos@athena.mit.edu (George W. Poulos (aka Cpt. Caveman)) (06/07/91)
--
George W. Poulos *Ad hoc, ad loc, - Jeremy *jiffy pop, nuclear
aka *and quid pro quo. Hillary *fission, i know
Captain *So little time. Boob *there's a connection
Caaavemaaaaaan!! *So much to know. PhD. *...there must be...gwpoulos@athena.mit.edu (George W. Poulos (aka Cpt. Caveman)) (06/07/91)
OK, let's try this again.
Hi I have a question on filtering out the 'herringbone' pattern that
appears on images when I use a b/w frame grabber to digitize the video
signal given by my (color) camcorder. I know that the color signal
frequency is ~3.5 MHz, and I built a trap filter tuned to that
frequency (It works OK). But I was wondering if there was any program
or algorithm for doing this _in software_. Any help would be
appreciated.
--
George W. Poulos *Ad hoc, ad loc, - Jeremy *jiffy pop, nuclear
aka *and quid pro quo. Hillary *fission, i know
Captain *So little time. Boob *there's a connection
Caaavemaaaaaan!! *So much to know. PhD. *...there must be...dave@imax.com (Dave Martindale) (06/08/91)
In article <1991Jun7.033634.27095@athena.mit.edu> gwpoulos@athena.mit.edu (George W. Poulos (aka Cpt. Caveman)) writes: >Hi I have a question on filtering out the 'herringbone' pattern that >appears on images when I use a b/w frame grabber to digitize the video >signal given by my (color) camcorder. I know that the color signal >frequency is ~3.5 MHz, and I built a trap filter tuned to that >frequency (It works OK). But I was wondering if there was any program >or algorithm for doing this _in software_. For NTSC, the frequency is 3579545 Hz, +- 10 Hz. Yes, you could build a digital notch filter. Dig out one of the standard texts on digital signal processing, and look up FIR (finite impulse response) filters. Basically, each pixel in the filtered image will be a weighted average of some span of pixels extending to the left and right of it on the same scanline. The number of input pixels needed to calculate each output pixel depends on the characteristics of the notch filter you want - how narrow and deep the notch, how much ripple there is either side of the notch. But you can get just about any filter characteristics you want if you're willing to do enough computation. Also, how is your frame grabber's clock synchronized? Does it just free-run, or is it locked to horizontal sync or colour subcarrier in some way? If it's locked to sync and stable enough, you can create a comb filter using the fact that the colour subcarrier phase inverts from one scanline to the next. Just average each pair of adjacent scanlines together, and the subcarrier plus all of its harmonics will disappear. (Note that this applies to two adjacent scanlines in the same *field*, if you have a frame digitizer). This does cause a loss of vertical resolution, but avoids the loss of horizontal resolution that the notch filter causes. It also results in a 1/2 pixel vertical shift of the image, but that doesn't matter much for black&white. If your frame grabber's sampling clock is stable enough, and if it digitizes the colour burst as well as the visible portions of the frame, you can actually extract *colour* images from the single- channel frame grabber output. The sampled colour burst lets you maintain a digital "colour subcarrier oscillator" that is phase-locked to the original, and you can demodulate the colour signal, filter to separate the chrominance and luminance, and then decode to get RGB. All in software. I've done this for the output of a digitizer where the sampling clock was phase-locked to three times the subcarrier frequency. That makes some of the decoding easier. The filters I used to separate luminance and chrominance are quite ugly, and produce visible ringing in the picture. But it does work. Someday I'll play with it some more and improve it, but I haven't touched it for about a year. Dave Martindale