Mosaic feedback for sketch training and retrieval improvement

The results of queries in image databases are usually presented as a thumbnail list. Subsequently, each of these images can be used for refinement of the initial query. This approach is however not suitable for queries by sketch: in order to receive the desired images the user has to recognise misleading areas of the sketch and to modify these appropriately. This is a non-trivial problem, especially for users with limited expertise in image retrieval and when complex features are used for the image description and comparison. Therefore, this paper presents a mosaic-based technique for sketch feedback, which combines the best sections of the database into a single image. An analysis of individual sections and the linked target images shows, which areas of the sketch lead to poor results and should be modified. Performance measurements show a significant increase of the recall rate.