An image database browser that learns from user interaction

摘要: Digital libraries of images and video are rapidly growing in size availability. To avoid the expense limitations text, there is considerable interest navigation by perceptual other automatically extractable attributes. Unfortunately, relevance an attribute for a query not always obvious. Queries which go beyond explicit color, shape, positional cues must incorporate multiple features complex ways. This dissertation uses machine learning to select combine satisfy query, based on positive negative examples from user. The algorithm does just learn during course one session: it learns continuously, across sessions. learner improves its ability dynamically modifying inductive bias, experience over Experiments demonstrate assist image classification, segmentation, annotation (labeling regions). common theme this work, applied computer vision, database retrieval, learning, building enough flexibility allow adaptation changing goals.