Stephen Berry wrote:

Paul Stuart wrote:
 Hi,
   Looking at the smooth routine in the dvsdk, it seems that de-interlacing is 
accomplished by taking the first video field and then expanding it by 2x.

 > From a comment in Smooth.c
     /*
      * Configure the IPIPE such that its input looks like a buffer half the
      * height of the input buffer, and consisting of every other line in the
      * input buffer.  Smoothing will be done by resizing this half-height
      * buffer to a full size frame.
      */


 So, if I'm understanding this correctly, the routine isn't so much 
de-interlacing the input, as it is throwing half of it away. Am I reading that 
right? If so, is there a way to smooth the image using the IPIPE and preserve 
full resolution?


You understand it correctly. Converting NTSC to a progressive frame
image with out motion artifacts has been a problem for a very long time.
The simplest method is to toss away one field and horizontally double
the other. This produces a whole host of visual artifacts.

well, this is actually a quite good method. It is almost impossible to
deinterlace properly if you can only work on one frame at a time. All
the really good deinterlacing methods use a "temporal" component by
looking at several consecutive frames (detecting moving vs static parts
etc)

For a single frame deinterlacer, all that I have seen so far was worse than
just throwing away one field and upscaling the other.


_______________________________________________
Davinci-linux-open-source mailing list
[email protected]
http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source

Reply via email to