Instead of casting the image as a zero, you might consider using im2double. Otherwise your image values may be very large causing a floating point problem.
You should check to see if any non-finite values ( see isfinite) are present in mu, sigma, P, or just Z. Since you are treating K as a vector, you could use std instead of std2 as well.
K = I(:);
P=normpdf(K, mu, sigma);
Z = norminv(P,mu,sigma);