This paper presents an object-independent approach to the structural analysis of images. Unlike object-specific methods, which reach effective solutions by using the specifics of the problems to be solved, the approach presented here is based on the most general limitations that characterize the spatial organization of the surrounding world (non transparency and local rigidity of the components of its objects, rules for their mutual screening). The limitations caused by the layout of the human visual system and of most computer-vision systems (affine or projective transformations that arise when an image is projected through a spherical lens) are also taken into account. Hierarchical structural methods of analyzing scenes and structural methods of analyzing textures are considered in terms of the approach presented here from single positions, taking into account the hierarchical organization of the surrounding world, in which objects can consist of sub-objects, while the textures on their surfaces can become independent objects of analysis when the spatial resolution of the image is increased. Based on the adopted limitations, an object-independent orthogonal basis of generalized structural elements is formed that is convenient for the structural decomposition of images at different hierarchical levels, and methods are developed for constructing structural descriptions based on such elements. Examples are shown of the analysis of actual images on the basis of the proposed approach. (C) 2008 Optical Society of America.
展开▼
