Relief Extraction of 3D Defaults on Specular Surfaces

Within the framework of the analysis of 3D textured environment through image analysis, we approach here the case of the rough surfaces with different photometric behaviour, for the analysis of the local variations of their relief. The studied surfaces are metallic and can include some local defaults which can be considered as having a Lambertian photometric behaviour. We propose here to detect, in a first step, the 2D localization of the default through image analysis. The localization method is based upon an unsupervised segmentation algorithm within the context of transferable belief model. Our iterative algorithm automatically segments the metallic plates onto regions. As the local defaults can be considered as macro-texture, our segmentation method succeeds to isolate this particular region of interest. In a second step, we propose an original adaptation of stereovision based on photometric model to the case of surfaces with different photometric behaviour. So, we clarify the incidence of these phenomena on the image grey level and then on the relief extraction technique. The suggested method is then used to characterize default with 2D resolution about 10times10 mum² and about 100 mum depth