Bilinear interpolation with non-aligned input points

Question

I have a non-grid-aligned set of input values associated with grid-aligned output values. Given a new input value I want to find the output:

      &nb

Answer 1

You can bilinearly interpolate in any convex tetragon. A cartesian grid is just a bit simpler because the calculation of interpolation parameters is trivial. In the general case you interpolate as follows:

parameters alpha, beta
interpolated value = (1 - alpha) * ((1 - beta) * p1 + beta * p2) + alpha * ((1 - beta) * p3 + beta * p4)

In order to calculate the parameters, you have to solve a system of equations. Put your input values in the places of p1 through p4 and solve for alpha and beta.

Then put your output values in the places of p1 through p4 and use the calculated parameters to calculate the final interpolated output value.

For a regular grid, the parameter calculation comes down to:

alpha = x / cell width
beta  = y / cell height

which automatically solves the equations.

Here is a sample interpolation for alpha=0.3 and beta=0.6

Actually, the equations can be solved analytically. However, the formulae are quite ugly. Therefore, iterative methods are probably nicer. There are two solutions for the system of equations. You need to pick the solution where both parameters are in [0, 1].

First solution:

alpha = -(b e - a f + d g - c h + sqrt(-4 (c e - a g) (d f - b h) +
        (b e - a f + d g - c h)^2))/(2 c e - 2 a g)    
beta  = (b e - a f - d g + c h + sqrt(-4 (c e - a g) (d f - b h) + 
        (b e - a f + d g - c h)^2))/(2 c f - 2 b g)

where

a = -p1.x + p3.x
b = -p1.x + p2.x
c = p1.x - p2.x - p3.x + p4.x
d = center.x - p1.x
e = -p1.y + p3.y
f = -p1.y + p2.y
g = p1.y - p2.y - p3.y + p4.y
h = center.y - p1.y

Second solution:

alpha = (-b e + a f - d g + c h + sqrt(-4 (c e - a g) (d f - b h) + 
        (b e - a f + d g - c h)^2))/(2 c e - 2 a g)
beta  = -((-b e + a f + d g - c h + sqrt(-4 (c e - a g) (d f - b h) + 
        (b e - a f + d g - c h)^2))/( 2 c f - 2 b g))