Fourier based estimation is shown to correlate more closely with the true fractal dimension than all other methods tested - this matches well with similar findings for the characterisation of one-dimensional fractals [14] . However, the results obtained here are not consistent with spectral density of the form f -(b+1) where b=2H+1 [12] , but instead suggest a spectrum of the form f -a where a=4H. Best correlation is obtained where tapering is not used and all Fourier coefficients are used equally in the curve fit process.
Estimation methods based upon an assumption of self-similarity (box counting, surface and blanket methods) perform poorly; this is likely to be a consequence of the self-similarity assumption being a poor approximation.
Continuing work includes the development of Fourier based measures of deviation from fractal behaviour - such measures may represent compact texture descriptors.