Analysis of Three Dimensional Textures Through use of Photometric Stereo, Co-occurrence Matrices and Neural Networks

This paper describes a novel methodology for automatic segmentation of three-dimensional surface textures (i.e. bump maps). The method could be employed in a range of fields where identification of changes in surface topography would be useful. The techniques involved include photometric stereo (PS) for capture of surface gradient data, with co-occurrence matrices (CM) being used for quantification of the surface texture. The resulting matrix values are modelled in terms of texture type, through use of a feedforward-backpropagation neural network (NN). A new surface can then be analysed, by calculating co-occurrence matrix values and consulting the NN for various random locations on the bump map. In this way the surface is automatically segmented into various texture types. The approach is believed to be novel in the way that the three-dimensional data, co-occurrence matrix and NN are combined. For example, co-occurrence matrices are usually applied to two-dimensional images, and modelled by reducing the matrix to parameters such as Energy and Entropy (with associated loss of information). The technique has been shown to be useful for segmentation of a range of texture types.