Purpose. In order to increase the precision and recall rate of image retrieval method, a new image retrieval method is proposed, which bases on the idea that different parts of an image have different discriminative capabilities and contribute to the image retrieval to different extents. The region of interest method was used to improve the effectiveness of the semantic mapping of content and image retrieval. Methodology. Based on the region of interest, we got a semantic class range in the feature space by semi-supervised learning. The semantic class range is the mapping of image and class. In view of the different image features, the most suitable distance measurement was chosen. Findings. The region of interest was divided from image by the improved Harris detection method and extract the low-level image features. The features in database were used as training data for semi-supervised learning to build the mapping of images and semantic classes which is updated iteratively by the relevance feedback. Originality. The map between an image region of interest and class labels was studied. Two kinds of distances were applied to calculate the different feature's similarity. The semantic mapping is updated in real time. The research realizes a novel and comprehensive image retrieval system. Practical value. Different needs of users were considered, and the very good experiment results have been received. The image retrieval method proposed in this paper is quicker and more effective than other methods.

  Якщо, ви не знайшли інформацію про автора(ів) публікації, маєте бажання виправити або відобразити більш докладну інформацію про науковців України запрошуємо заповнити "Анкету науковця"