Data fusion and label weighting for image retrieval based on spatio-conceptual information

Using as experimental platform an image retrieval method based on a spatio-conceptual representation of images, in this paper we investigate two main concerns on annotationbased image retrieval: label weighting and data fusion. On the one hand, we analyze the influence of different weighting schemes on the quality of the retrieval performance, and, on the other hand, we study the application of fusion techniques for queries represented by more than one sample image. Particularly, we aim to compare fusion methods based on score information and on ranking information in order to determine the most adequate approach for the annotation-based image retrieval task.