Research On Ontology Based Image Retrieval On Hadoop Cloud Computing

With the coming of the era of big data, image data in each network node have increased dramatically, and images are more and more complicated, which gives people great convenience, but also brings great burden to the massive images retrieval. How to effectively retrieve such as image and other multimedia data becomes a hot and difficult spot to research. In the existing methods of image retrieval, the text based image retrieval has the problems of strong subjectivity and incomplete expression of the meaning of the image; the content based image retrieval has the semantic gap between the low level semantic and the high-level semantic; the semantic based image retrieval is based on a single concept, so it does not take the relationship between concepts into account. Therefore, how to retrieve the target image effectively from the large scale image database is an important problem to be solved.The semantic web makes up the semantic gap between low-level features and high-level semantics. As the semantic layer of the semantic web, ontology structured and conceptualized knowledge representation. As a technology that can handle and store large data, Cloud computing has the advantages of high reliability, scalability and low cost. This paper focuses on ontology based image retrieval method on Hadoop Cloud Computing, and establishes high recall and precision image retrieval system.Based on the problem above, this paper proposes an image retrieval system based on ontology. Firstly, we segment images using FCM algorithm with spatial information, and use clustering algorithm to cluster large image elements. Secondly, the ontology of sports field is established, which is used to label the images of the field;Then, the traditional CMRM is adopted in the training images to obtain the basic image annotations;Finally, the graph learning is applied to refine the basic annotations based on ontology concept similarity. The top N keywords in the probability table are chosen as the final annotation results.Experimental results have shown that the proposed image segmentation algorithm incorporating spatial information can segment images better, and the domain ontology is introduced to image annotation process, which defines the relationship between image semantic concepts and can not only make full use of low-level features of images, but also conform to the visual understanding of the images. The recall and precision of image retrieval are improved.