The Research On The Automatic Image Annotation Based On Ontology

With the rapid development of science and technology, digital image has becomeone of major resources in the field of computer. The traditional method of manualimage annotation cannot satisify the increasing needs of people as its problems ofstrong subjectivity and huge burden, so how to automatically generate annotationwords for image has become a serious problem. The goal of approach of automaticimage annotation is to build a bridge between the low-level features and high-levelsemantic, and to construct an automatic annotation framework to generate vocabularyto describe the semantic in image. With the development of the theory and relevanttechnology in machine learning, statistic method and semantic web, many scholarshave proposed many methods in automatic image annotation and adopted ontology todescribe the content of the image. In this paper, we proposed the method of automaticimage annotation based on ontology. The paper has further research in relatedproblems of automatic image annotation, the main work in the paper are as follows:Base on the review of the proposed methods of automatic image annotation, weintroduced relevant theory and knowledge of image semantic, including the semantichierarchy structures of image semantic, the description methods of image semantic,extraction low-feature of the image and the related theory of ontology technology.According to the existing methods of extraction region of interest, we proposed anew method of image description based on the interest of region. Firstly segmentedthe image into meaningful image regions with the improved clustering algorithm, andthen calculated the degree of interest for each region according to the variance of grayand area ratio of region, and then compute the visual weights to measure theimportance for each region in image, and also do the experiment in image retrievalbased on that method of regions of interest. Experimental results show that themethod of description image based on regions of interest proposed has better retrievalperformance.Based on the analysis of the methods of image annotation based on ontology, weproposed an image method of automatic image annotation based on ontology.Automatic annotation process is divided into two steps: object semantics extractionand scene semantic extraction. The visual weight for image region was applied intothe training set in image annotation, and combining Bayesian methods to compute priori probability and posterior probability of object concept for region and obtain theobject semantics of the image, and then according to the obtained object semantics,combining with the relationship between objects and scene in ontology and theassociation between objects and scene in the training set, we can extract scenesemantics of the image. According to the experiment of image retrieval based onsimply word query and extended query based on ontology, it proved the methods ofautomatic image annotation base on ontology we proposed has better performance.