If you can manage to transform the cdf plot into two arrays, one the discrete variable values, 'v', and the second the corresponding cumulative probability values, 'p', of these values, then you can use matlab's 'trapz' function to find the distribution's mean and standard deviation values from these arrays. The mean value, 'm', would simply be the integral of 'v' with respect to 'p' and the standard deviation would be the square root of the integral of (v-m)^2 with respect to 'p'.
I am not very knowledgeable about tracing plots to create discrete arrays, but I notice at least one File Exchange routine that appears to undertake such a task. It is located at:
http://www.mathworks.com/matlabcentral/fileexchange/3832-matlab-plot-tracing-utility/content/GTrace.m
There may be more in the File Exchange or even in one of Mathworks' toolboxes.
