VISTO: A new CBIR system for vector images

摘要

In this paper, we present the main features of VISTO (Vector Image Search TOol), a new content-based image retrieval (CBIR) system for vector images. Though unsuitable for photo-realistic imagery, vector graphics are continually becoming more advanced and diffused. Vector images are fully scalable, resolution independent, not restricted to rectangular shape, allowing layering and editable/searchable text. Notwithstanding this increasing interest, the research area concerning CBIR systems for vectorial images is quite new, and our research on a vector based CBIR system actually derives from a precise request of vector based application experts that did not find appropriate solutions to their retrieval problems in customary shape-based CBIR system. To the best of our knowledge, VISTO is the first CBIR system for vector images proposed in the literature, and it supports the retrieval of images in SVG (scalable vector graphics) format.rnWe discuss VISTO from both the engine and the interface points of view, and then evaluate its engine from an experimental point of view within an advanced high quality 2D animation environment supporting cartoon episodes management. To satisfy the requirements of stroke-based applications, in VISTO vector images are modelled as inertial systems and then associated with descriptors representing visual features invariant to translation, rotation, and scaling transformations. Furthermore, to efficiently serve different application domains, the engine offers a variety of moment sets as well as different distance functions for similarity computation. The graphical interface offers tools that help in the selection of criteria and parameters necessary to tune the system to a specific application domain.