DSPRelated.com
Forums

please recomend algorithm for image surface interpolation?

Started by walala March 21, 2004
Dear all,

I am thinking of treating nature image like surfaces, suppose somehow there
are some missing pixels in this image, and I want to do surface
interpolation to reconstruct the missing pixels. Which surface interpolation
algorithm can do this job best? I hope this surface interpolation approach
will be a lot better than bilinear interpolation, or separable
interpolation, am I right? Any enlightening
ideas/thoughts/suggestions/comments will be greatly appreciated! Thank you
very much!

-Walala


Walala,

If I understand you well, I had a similar problem.
I need to process data from an interferometric surface profiler. The "steep"
parts of the surface usually do not yield valid pixels. That's about 30 or
40% invalids in my image.
Blindly using griddata to interpolate them takes too long for my ca.400x700
images.
I reduced the amount of work to do for griddata by first determining the
perimeter of the missing pixel regions, using the perimeter values as the
known values for the interpolation, and using only the missing pixels as the
values to be interpolated.
That runs in about a minute or so.
The (proprietary) software supplied with the equipment does the
interpolation a lot faster, so there must be better way.
I hope this helps, but I am anxious to see better replies to your posting

Patrick


walala wrote:

> Dear all, > > I am thinking of treating nature image like surfaces, suppose somehow > there are some missing pixels in this image, and I want to do surface > interpolation to reconstruct the missing pixels. Which surface > interpolation algorithm can do this job best? I hope this surface > interpolation approach will be a lot better than bilinear interpolation, > or separable interpolation, am I right? Any enlightening > ideas/thoughts/suggestions/comments will be greatly appreciated! Thank you > very much! > > -Walala
Try Bspline interpolation (e.g. the Multi-level Bspline Approximation MBA algorithm) for 2d or fit a Radial Basis Function (RBF) with a thin-plate spline basic function. That should do what you want.
"walala" <mizhael@yahoo.com> wrote in message
news:c3j79b$4dr$1@mozo.cc.purdue.edu...
> Dear all, > > I am thinking of treating nature image like surfaces, suppose somehow
there
> are some missing pixels in this image, and I want to do surface > interpolation to reconstruct the missing pixels. Which surface
interpolation
> algorithm can do this job best? I hope this surface interpolation approach > will be a lot better than bilinear interpolation, or separable > interpolation, am I right? Any enlightening > ideas/thoughts/suggestions/comments will be greatly appreciated! Thank you > very much! > > -Walala > >
In fact I have more question on this: I tried some surface interpolation function, the problem is that they do not use boundary conditions: for example, I need to interpolate some blocks in image, each block has some pixels missing, and some pixels existing, suppose I now have 2 adjacent such blocks need interpolation. If I just do independent interpolation on each block, the boundary between the two adjacent blocks after individual interpolation will be very bad/discontinuous. I want to ask is there an interpolation algorithm that can simultaneously interpolate some adjacent blocks and keep boundary smooth to other adjacent interpolated blocks? Thanks a lot, -Walala