Most AFMs have a software that outputs the height-height correlation. There are three standard classes within which the majority of h-h correlation functions fall - self-similar, self-affine and liquid-like. Of these the liquid-like has infinite correlation length and the others have lateral correlation lengths within their expressions. These are given in standard references (even in Wikipedia). Import your experimental h-h correlation to origin and fit with these functions to get a best fit and the correlation length.
Or alternatively use software like SPIP (http://www.imagemet.com/)
In addition to wha Prof. Datta mentioned above, it is important to check which height-height correlation function is suited for the system you are investigating - the analysis will always provide some numbers for the correlation length, the Hurst parameter and the roughness, but the question is if those values have a physical meaning!
You could use the freely available software Gwyddion for plotting the Height Height correlation or Autocorrelation function for the AFM images and from that calculate the lateral correlation length. Additionally, this software directly gives you the extrapolated value of the correlation length using it's in built algorithm.