Conclusions

We have proposed here a way of quantifying 3D surface roughness. The roughness is expressed with the help of a 2D co-occurrence matrix of lengths of surface chords with the angle between the surface normals associated with them at their two ends. Various methods of extracting measures of surface roughness from these histograms have been proposed and used to characterise the surfaces of colons from patients suffering from colitis. The classical measures of histogram similarity/dissimilarity (eg test, correlation coefficient etc) as well as a new proposed feature that characterises the roughness expressed by each histogram correlate well with the class of each object and its ``distance'' from the most compact class in the image database we have, namely the class ``colitis sufferer''. However, non of the measures on their own is enough to identify the class uniquely. This is not surprising as usually clinical diagnosis is not based on a single feature of a condition. The roughness of some brain images was also computed and the result qualitatively agreed with the already known fact that the human brain is rougher than that of a monkey. These results are sufficiently encouraging, but rather inconclusive in establishing the proposed method as an efficient tool for object identification. More experimentation is needed with a more appropriate 3D image database.

Acknowledgements: A Royal Society grant supporting this work is gratefully acknowledged.