Resampling method for orthophotos

[timwhittome]

A client has asked what resampling method is used when generating orthophotos. The previous software I used for orthorectification (Geomatica) gave several choices - nearest neighbour, bilinear interpolation (weighted average of four closest pixels), cubic convolution (weighted average of 16 closest pixels) and a host of other more esoteric filters. I generally used cubic convolution as giving the best-looking result. What does PS use? There are hints that for some purposes it uses nearest neighbour, but I can't find any information specifically about generating orthophotos.

[Alexey Pasumansky]

Hello timwhittome,

PhotoScan uses biliniar interpolation during orthomosaic generation process.

[timwhittome]

Thanks Alexey. Another question - how does PS calculate the dense cloud at lower resolutions than Ultra High? For example, at High does the program look at every second pixel (or fourth pixel for Medium etc) or does it reduce the resolution of the photographs by successive factors of 2 and use averaged values for every pixel in the reduced images?And would this make a difference to the accuracy of resulting DEMs?

[Jimbob]

It would be great if a future release of Agisoft allows for orthophoto output in tiff format in alternative sampling types (bilinear, bicubic, nearest neighbor, etc). I think nearest neighbor gives the best results in other geospatial software (inpho, geomatica, etc).

I my opinion bilinear looks soft and blurry. Nearest neighbor is a better option!

Jamie