Re: Back to basics: Interpolation teaser

"If one zooms in at the result image, the response appears different
for the dots away from the center as compared to at the center.

Do you also notice that?"

Yes, for all integer scale factors, the responses should be the same. For
non integer scales, they change with location. Sometimes programmers cheat
by enlarging slightly larger than specified to avoid extrapolation near the
image boundaries, but, in this case, if anything, the image appears
slightly smaller than it should.

Also, the sum of luminances of the response pixels should equal the
scale^2 times the luminance of the impulse. For the center response, I
count 1540 when it should be 4*255 = 1020. Possibly, you have clipped
negative values to zero that would account for this. That is why you
should look at the response against a grey background.

"I wonder if that is because of floating point precision error in mapping
co-ordinates on to the non-integral source image grid."

No, 32 bit floating point is adequate.

.

Relevant Pages

  • Re: while situations barely melt teachers, the negligences often run for ever the historic reluctanc
    ... Gavin, still floating, rates almost silently, as the wheel approachs ... unless Mohammed conceives symptoms for example Usha's ... tap the minor bottle and merge it in response to its catalogue. ...
    (sci.crypt)
  • Re: Low Pass Filter Design Help
    ... Is it still considered FIR when implemented with a delayed ... differentiator (delay sets the window size) followed by an ... principles of floating point comparison. ... But it is an interesting response, ...
    (comp.dsp)
  • RE: Adding floating point emulation support for MIPS R4300
    ... Thanx for your response. ... Since I already have an image created for a device with FPU, can I make any changes to the existing image to enable this functionality? ... included script samples are subject to the terms specified at ... has a floating point unit, however, I want to use floating point emulation. ...
    (microsoft.public.windowsce.platbuilder)
  • Re: printing double value
    ... >Because a floating point number is in a reasonably complicated format that ... his boss came in and found him doing a classic boiling point experiment. ... The boss asked him why he was doing that. ... response was 'In a jam jar on the ...
    (alt.comp.lang.learn.c-cpp)
Quantcast