On Jul 2, 11:04=A0pm, "Qingeng.Deng" <Qingeng.D...@[EMAIL PROTECTED]
> wrote:
> exactly speaking, this impulse response ****tion is used to model the
> tem****al response of human vision as looking a motion object. So, when
> i do my master thesis about video processing, I don't know how to
> implement this filter. Generally, the sigma is equal to 160ms, and tao
> is 0.2.
> =A0 =A0 As t>200ms, this impulse response is close to zero. Then it is
> sampled =A0with the sampling period equal to frame rate, e.g. 50Hz. Is
> this method correct ?
one thing to worry about is what the spectrum of this, err, "thing"
is. (i can sorta visualize it, starts at 0 gets to the max when
t=3Dtao, decays back to 0 and sigma is a sorta shape parameter.) since
it's expressed in continuous-time and has a sharp edge (maybe it's not
so sharp), you can generally expect that the spectrum goes up to
infinity. you have to have a sampling rate sufficient to get most of
it.
the, i dunno, window the impulse response and sample it. but you'll
need a ****uva lot more than 7 or 8 taps. i think, if you're trying
to a real-time image processor and the computational bandwidth of if
gives you only 7 or 8 taps, you're screwed. sure, you can sample your
continuous-time target, but what you will get in the frequency
response is your ideal but aliased in the frequency domain about the
Nyquist. to what degree you can tolerate that determines the sam[ling
rate. and to what degree you can tolerate sharp transitions in the
frequency response getting less sharp will then determine how many
taps you need. (this is assuming a rectangular window.)
r b-j
|
