Shape Extraction by Nonlinear Diffusion
Ebroul Izquierdo · 2001
An advanced scale-space technique for image segmentation and pattern recognition in computer vision is presented. The approach is based on the Perona-Malik [1] nonlinear diffusion model. The idea at the heart of this technique is to smooth the image within object boundaries, inhibiting diffusion across the contour and even enhancing the contrast along the boundaries. For general image simplification and segmentation, the Perona-Malik paradigm leads to impressive results outperforming clearly well-established filters like the canny operator and morphological schemes. Nevertheless it cannot be used directly to carry out the more complex object segmentation task. In this context an extension of the diffusion model is introduced to extract shapes of complete physical objects present in the scene. In the new formulation information about the structure or dynamic of the scene is used. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.