I was performing upscaling via OpenCV's bicubic interpolation algorithm.
Essentially, I want to scale an image to a size WxH.
The initial images have sizes W/2 x H/2 and W/4 x H/4 for x2 and x4 upscaling respectively.
I also timed the algorithms for several images and averaged out the results and it appeared that the x4 upscaling is slightly slower than the x2 upscaling.
Why is this so? As I understood it, bicubic interpolation is performed via convolution, in which case the output image is the same size. So shouldn't the algorithm take the same time?
If the argument is that the x2 has four times the data that that is simply copied to the output rather than worked out, then shouldn't x4 upscaling take four times as long (this is not the case)?
Thank you.