New wavelet based space-frequency analysis methods applied to the characterisation of 3-dimensional engineering surface textures
The aim of this work was to use resources coming from the field of signal and image processing to make progress solving real problems of surface texture characterisation. A measurement apparatus like a microscope gives a representation of a surface textures that can be seen as an image. This is actually an image representing the relief of the surface texture. From the image processing point of view, this problem takes the form of texture analysis. The introduction of the problem as one of texture analysis is presented as well as the proposed solution: a wavelet based method for texture characterisation. Actually, more than a simple wavelet transform, an entire original characterisation method is described. A new tool based on the frequency normalisation of the well-known wavelet transform has been designed for the purpose of this study and is introduced, explained and illustrated in this thesis. This tool allows the drawing of a real space-frequency map of any image and especially textured images. From this representation, which can be compared to music notation, simple parameters are calculated. They give information about texture features on several scales and can be compared to hybrid parameters commonly used in surface roughness characterisation. Finally, these parameters are used to feed a decision-making system. In order to come back to the first motivation of the study, this analysis strategy is applied to real engineered surface characterisation problems. The first application is the discrimination of surface textures, which superficially have similar characteristics according to some standard parameters. The second application is the monitoring of a grinding process. A new approach to the problem of surface texture analysis is introduced. The principle of this new approach, well known in image processing, is not to give an absolute measure of the characteristics of a surface, but to classify textures relative to each other in a space where the distance between them indicates their similarity.