Ontological shape-description, a new method for visual information retrieval

There is a growing need for efficient visual information retrieval systems which take into account particular features of images, in order to avoid nonsensical results. We propose a new method for content-based image retrieval, which can be divided into two main parts: 1) automatic segmentation and extraction of shapes from image sub-regions; 2) ontological descriptions of shapes contained in the images. Our method allows users to perform sketch queries; these are simple drawings which represent the main idea of something the user is expecting to retrieve from the results. Once the user introduces a sketch, it is simplified using discrete curve evolution; then it is turned into a tangent space representation. Similarity measures based on turning functions are used to recover similar images. Ontology reduces the proportion of nonsensical results since it permits the recovery of images related semantically, even if they do not have similar shapes.