Method for automatic creation of a database of images searchable by its semantic content

摘要

Is performed a plurality of segmentations (131) an initial image (4) in homogeneous components using different criteria to identify and isolate several series of objects (Ai, Bi, Ci, ...) consisting of thumbnails segmented (41, 42, 43) associated with geometric descriptors of bounding boxes of those objects (Ai, Bi, Ci, ...), there is provided a fusion of the results of segmentation processes are characterized by numerical values ​​each of thumbnails object resulting from a melting and comparing each of the thumbnails with the indexed thumbnail images of a reference data base (12) comprising a set of textual descriptions of everyday elementary objects. It provides with relevance rate list of textual descriptions of everyday basic objects consistent with the object thumbnails resulting segmentation and merger.