[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: HiRes color correction - Nasty Details
>My (limited ) understanding of the hardware approach is that although the
>resolution may be very high, it cannot be truly resolution independent. I would
>be grateful if Steve could elaborate on this.
There really is nothing magic about resolution independance. Most colour
correction algorithms process either individual pixels or small groups of
closely related pixels. This means that the essential DSP hardware only
has to be fed a few pixels at a time in order for it to complete the
As in the software approach there is a direct trade off between processing
time for the full frame compared with the overall size of the frame.
The key to the whole system is in the move from the TV world to computerland.
When processing TV signals Sync pulses arrive with relentless precision.
Every calculation must be completed exactly at the right time before the
next sync pulse arrives.
With computer data we have a handshake system which gives the DSP hardware
the ability to say 'hang on a minute, I'm not quite ready'.
Obviously the hardware must be capable of processing the required amount of
data within a given time period but this can be done in bursts --- all the
easy calculations may be performed very very quickly leaving more time for
the more esoteric functions. More importantly a system capable of
processing one resolution in real time [[by which we mean usually 25or 30
frames a second] ] can also process a image in much higher resolution but
at a proportionally slower rate.
For example a single MegaDEF channel has an average throughput of around
160Mbytes per second. Depending on the interfaces that we provide this
means that this unit can simultaneously process 2 or 3 standard 625 line
images [approx 50MBytes per sec] in real time at OR a 2000 line image
[approx 350Mbytes per sec] at about 12 frames per second --- very much
faster than any software based solution. The design of the MegaDEF allows
for upto eight modules to be used in parallel to provide very high
processing rates where needed.
Hope this all makes sense !!!!!!