Methods Based On Concept Lattice Image Annotation Semantic Level

With the rapid development of Internet technology and the popularizationof digital equipment, large image databases have also been produced. In orderto effectively manage and mine valuable information of databases, imagesemantic annotation has become the urgent demand. BOV model is a majorrepresentation of image content, which has been widely used in image semanticannotation. Effects of visual dictionary size, semantic gap between low-levelfeatures and high-level semantic and polysemy in the middle-level semantic onclassification performance are currently active and challenging problems in theresearch field of automatic annotation. Concept lattice is an effective tool fordata mining, concept hierarchy between the nodes is helpful to extractdifferent capacities and different levels of knowledge. In this paper, conceptlattice is used as a classification model for studying the method of imagesemantic hierarchical annotation. The main research works are shown asfollows:1） A method of visual dictionary is presented by analyzing the conceptlattice hierarchy. First, the initial visual dictionary of training images on BOVmodel is generated. Then, with the use of concept lattice’s hierarchy analysis,the different granularities of reduced visual dictionaries are extracted from theconcept lattice by setting different extension thresholds. Finally, the polysemy isdeleted by making XOR operations on all types of the reduced visualdictionaries, and a visual dictionary for better describing the image content isgenerated. Experimental results show that this method is effective.2）A method of image semantic hierarchical annotation is presented basedon concept lattice. Firstly, image scene semantic and object semantic are dividedinto the high-level semantic and middle-level semantic by according to thehierarchical description of image semantic, and the concept lattice based on aformal context of BOV model, object semantic and scene semantic isconstructed. Secondly, with the use of concept lattice’s hierarchy analysis, the hierarchical annotation models are established by setting different extensionthresholds. Then, hierarchical annotation of the test images is implemented bythis method. Finally, Experiments show that the method can effectively carry outthe image semantic hierarchical annotation.3） On the basis of the above research， the concept lattice based imagesemantic hierarchical annotation prototype system is designed by usingMATLAB and JAVA as development tools.