• Scale factor. The dimensions of the interpolated image are determined by the scale factor.
For example, interpolation with scale factor of 4 increases a 100×100 image to
400×400. For non-integer scale factors, the dimension is rounded up.
• Deconvolution strength σh. The interpolation finds an
image satisfying the degradation model
input = ↓(h ∗ interpolation)
where ↓(⋅) denotes subsampling and h is a Gaussian with standard deviation
σh in units of input pixels. The σh parameter
controls the deconvolution stength of the interpolation. Set σh = 0
for no deconvolution.
• The algorithm can run in two different ways:
Interpolate image
Coarsen, interpolate, and compare
“Interpolate image” directly interpolates the selected image.
With “Coarsen, interpolate, and compare,” the image is coarsened to create the input
image according to input = ↓(h ∗ original).
The coarsened image is then interpolated and compared with the original image.
• Source code for this demo:
zip
tar.gz
You can run the algorithm in two different ways.
Or you can run it after selecting a subimage by clicking two
opposite corners of the subimage.
IPOL Journal · Image Processing On Line
