I don't know anything about images. In general, to calculate the standard deviation related to a histogram of data in a variable X, you would apply std, normfit, or dfittool to the X, not to the histogram data. I hope that's what you did.
I'd expect the three to give the same answer in that case. I'd expect the mle function to give a slightly different answer, since it compute the maximum likelihood using a 1/n factor, rather than an unbiased estimate based on a 1/(n-1) factor.
