An automated surface texture characterization system

[Truncated abstract] The development of an automated surface characterization system/method is of great importance to engineering and also to medicine. Such system, once developed, would be able to replace human experts engaged in quantifying the surface textures and assist medical personnel in the diagnosis and prognosis, for example, of joint diseases. The system would also improve the efficiency, reliability, accuracy and also reduce costs of monitoring or diagnosis. Various methods of 3D surface characterization have been reported in literature but they only seem to perform well when applied to isotropic surfaces at a single scale. However, real biological or engineering surfaces are not only anisotropic but also multiscale objects. Therefore, there is an urgent need for the development of a new system that characterizes surfaces at different scales, i.e. in a multiscale manner. This thesis is divided into three parts. In the first part, three new surface characterization methods were developed and their accuracy, in measuring surface roughness and anisotropy, was investigated using fractal surfaces and X-ray bone images. The newly developed methods are: a fractal signature Hurst orientation transform (FSHOT), a variance orientation transform (VOT) and a blanket with rotating grid (BRG). Unlike other methods, they have a unique ability to characterize surfaces at different scales in all possible directions. The results from this research have demonstrated that, amongst methods tested, the VOT method is most reliable and accurate. This part of the thesis is described in detail in Paper 1. The second part contains a series of works in which the performance of the VOT method was evaluated using medical and engineering surface images. II Databases of trabecular bone (TB) texture images from osteoarthritic (OA) and non-OA subjects were constructed.