Automated annotation system for natural images

Automated annotation of digital images remains a highly challenging task. This process can be used for indexing, retrieving, and understanding of large collections of image data. This paper presents an image annotation system used for annotating natural images. The proposed system is using an efficient annotation model called Cross Media Relevance Model for the annotation process. Image´s regions are described using a vocabulary of blobs generated from image features using the K-means clustering algorithm. Using SAIAPR TC-12 Dataset of annotated images it is estimated the joint probability of generating a word given the blobs in an image. The annotation process of each new image starts with a segmentation phase. An original and efficient segmentation algorithm based on a hexagonal structure is applied to obtain the list of regions. Each meaningful word assigned to the annotated image is retrieved from an ontology derived in an original manner starting from the hierarchical vocabulary associated with SAIAPR TC-12 and from the spatial relationships between regions.