Image Querying

Image querying refers to the problem of finding, within image databases (Image DBs), objects that are relevant to a user query. Classical solutions to deal with such problem include the semantic-based approach, for which an image is represented through metadata (e.g., keywords), and the content-based solution, commonly called content-based image retrieval (CBIR), where the image content is represented by means of low-level features (e.g., color and texture). While, for the semantic-based approach, the image querying problem can be simply transformed into a traditional information retrieval problem, for CBIR more sophisticated query evaluation techniques are required. The usual approach to deal with this is illustrated in Fig. 1: By means of a graphical user interface (GUI), the user provides a query image, by sketching it using graphical tools, by uploading an image, or by selecting an image supplied by the system. Low-level features are extracted for such image; such features are then used by the query processor to retrieve the DB images having similar characteristics, with the intuition that images having similar content also share similar features. How the set of relevant DB images is determined depends on which low-level features are used to characterize image content, on the criterion used to compare image features, on how DB objects are ranked with respect to the query (based on either a quantitative measure of similarity or qualitative preferences), and, finally, on whether the user is interested in the whole query image or only in a part of it. All these aspects strongly influence the query evaluation process.