Mapping an image to a surface

Hello,

I am trying to write a function that will keystone an image in a very specific way. Essentially, I need the image to be manipulated so that, if x and y were to represent the width and height (relative to the bottom) of the image, then the image would be mapped to the surface

z(x,y) = 0 with bounding regions

(bottom of image) < y < (top of image)

l(y) < x < r(y).

l(y) and r(y) are two different logarithmic functions (l(y) has a positive correlation while r(y) has a negtive). The image maintains its height but slowly shrinks in width the further up the image, making sort of a curve sided trapezoid.

I have attempted using the warp function, using the image boundaries as the x, y, z inputs, however this did not work. I have looked into also making a custom tform but it looks as though it can only translate pixels and not stretch or compress them. However, I am still relatively new to matlab and could be mistaken. All I need is just a general point in the right direction or some suggestions.

thanks in advance for any consideration or response,

mike

Answer 1

[Walter Roberson]

As a first pass:

outIM = zeros(size(IM));
h = size(IM,1);
w = size(IM,2);
for Y = 1:h
  L = round(l(Y));
  R = round(r(Y));
  p = round(interp1(linspace(L,R,w),1:w,1:w));
  pv = ~isnan(p);
  outIM(Y,p(pv),:) = IM(Y,L:R,:);
end
outIM = flipud(outIM);

This is far from ideal, as it pulls the values from individual complete pixels rather than some weighted color; likewise because it uses ragged edges rather than anti-aliasing the edges. Better yet would be a 2D interpolation.

I think it might be possible to vectorize the linspace and interp1 parts, but not the filling part, at least not with this quantized weighting.