|
|
|
First, the result! I didn't implement any denoising pictures. I started with these three images:
|
|
|
|
And ended up with this image. This is all 3 images averaged together, without any denoising, blending or other masking. The center bit is just the three images raw added together and divided by 3
I used a slightly different method to calculate A than the one discussed in class.
Instead of modeling our transformation matrix as x, the origin points as a matrix A and the reulting points as results B, I differentiated the least squares equation with respect to A
The end result I got was the solution to the least squares equation for A was as follows, where xi is the origin point in the first image, and bi is its matching point in the other image. We are using homogeneous coordinates, though that doesn't change the equation.
A = (sum(b_i * x_i.T) * inv(sum(x_i * x_i.T)))
Heres a warp of the sidewalk to flat
