🚀 go-pugleaf

In article <1d4bptINN7rr@elmer.ctd.comsat.com> rao@ctd.comsat.com (Ashok Rao) writes:
>In article <1d3ji1INN759@cs.tut.fi>, defee@tukkasotka.tut.fi (Irek Defee) writes:
>|> I would like to know present state-of-the-art in algorithms for
>|> high-quality image z o o m i n g, ...
>Scaling down is easy, just low-pass filter and decimate. 

I think it just seems to be easy since the linear approach would require in
fact complicated multirate conversion and decimation filters for each
zooming ratio. That would lead to blurring artifacts. 

However, there is even more intricacy in this problem. This can be
illustrated on the following example. Let's have an image which is
just a cosinusoid of certain (enough high) frequency. Using the
standard approach with lowpass filter and decimation, this cosinusoid
will be removed and the resulting image will be perceptually
unacceptable as a zoomed-down version of the original one. In this 
case, a better version wolud be a cosinusoid with frequency scaled to
the size of the new image.

>Scaling up on computer
>displays is usually done by just repeating pixels. Better results can be 
>obtained by bilinear interpolation with some loss in resolution however. Good
>results can be obtained by interpolation followed by a good (many taps) 
>low-pass filter. 

Here again blurring will result. Some nonlinear technique might be
better, as we don't need to interpolate e.g. step edges.

>-Ashok Rao (Comsat Laboratories)

Thanks,

Irek Defee