Segmentation of Natural Images Using Scale-Space Representation with Multi-Scale Edge Supervised Hierarchical Linking

In general purpose computer vision systems, non-supervised image analysis is mandatory in order to achieve an automatic operation. In this paper a different approach to image segmentation for natural scenes is presented. Scale-Space representation is used to extract the structure from meaningful objects in the image. A hierarchical decomposition of the image is performed from the iso-intensity paths. The Scale-Space stack is generated using isotropic diffusion on the basis of linear Scale-Space theory. From that, the independence of the algorithm from the image content and particular characteristics is ensured. In the framework of this work, it is also introduced the use of additional information to improve the robustness in the structure extraction. In addition to the set of several diffused versions of the image, a representation of edges through scale is included as a feature in order to supervise the generation of the hierarchical tree that represents the image.