PSF accuracy measure for evaluation of deblurring algorithms

Computes an expected MSE between images deconvolved with the estimated PSF versus the ground truth PSF.

Computes a theoretical MSE between images deconvolved by the estimated PSF and the ground truth PSF. No real deconvolution is performed. The error is the expected MSE if the deconvolution were performed on images with the given power spectrum.

The first two input arguments are images of the estimated PSF and the ground truth PSF.

The first output argument is the expected MSE between the GT images and the estimated images if the estimated PSF was sed. The MSE depends on the content of the image, so additional arguments must be provided to the function.

The third input argument (S) is the power spectrum of images abs(fft2(img)).^2., e.g. the average of all test images or an artificial power spectrum of the form 1/f^p (f .. distance from the 0,0 frequency, p ... parameter).

The fourth input argument (nsr) is the noise level. It is important to specify a realistic value. It can be either a full noise power spectrum or a noise standard deviation, depending on the fifth argument (mode). Typical values are nsr=std_of_noise and mode="sigma".

The sixth input argument (do_shifts), if true, shifts the input PSFs to different positions to find the lowest MSE. This is necessary if the PSFs are spatially aligned.

For more instructions consult the help in the m-file.