You can't -- at least not directly.  Free energy surfaces are undefined up to a constant shift (that is, you can set the "zero" free energy to whatever value you want, and only free energy differences have any meaning).

A common practice is to set the minimum free energy (that is, the most populated "bin") to zero (which is what seems to have happened for your data).  The free energy for each bin in your surface, then, is computed as

G(phi, psi) = -kT*ln(hist(phi, psi)/MAX)

In the above equation, G is your free energy, hist(phi, psi) is the value of your histogram (bin) at that particular phi/psi value, and MAX is the population of the lowest-free-energy structure.  According to your file, the only known values are G(phi, psi), k, and T.  As a result, the best that you can compute is the ratio of the populations of each bin to the bin with the largest population.

Hello Jason, Neha.  Hair-splitting aside it seems like what's needed is an easy way to parse-out the free energy surface into basins and integrate them.

Depending on the noise in your data Neha you might want to smooth it in order to define minima, maxima and basin boundaries.  scipy.interpolate.RectBivariateSpline can do this for you.  With some fiddling it should return a bivariate spline.  You can change the resolution by changing the smoothing parameters and the order of the spline (cubic, quartic etc).   Gradients of the resulting continuous surface should be well-defined such that you can precisely specify minima, saddle points and basin boundaries.

My opinion is that for physical purposes a basin barrier less than 0.5 kBT in height is not very important (time to cross drops to less than nanoseconds), but if you want to really go to town on your histogram bins you can try one of these free-energy-disconnectivity graphs from Wales' lab, he has many examples on his homepage: http://www-wales.ch.cam.ac.uk/