基于显著性分析的标签排序.docVIP

 Adaptive Tag Ranking based on Saliency Analysis# ZHAO Ripeng, SONG Zehai, FENG Songhe** (School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044) Abstract: Tag ranking has become a hot research topic due to its importance for image analysis and 5 10 15 20 25 30 retrieval. Existing annotation methods about tag ranking can be roughly classified into two categories: tag relevance ranking and tag saliency ranking. Both methods have pros and cons. In this paper, we propose an adaptive tag ranking based on saliency analysis which combines the advantages of tag relevance ranking and tag saliency ranking. The main idea behind the approach is apparently simple. In short, to the given image, we first carry out image salient region detection and image saliency analysis by machine learning techno-logies like Support Vector Machine (SVM). If there exist visually salient regions of the given image, the corresponding annotated tags can be ranked according to the saliency property of the corresponding visual content; else tags can be ranked according to the relevance scores to the content of the image. The performance will be better than existing methods. To demonstrate the effectiveness and efficiency of the proposed algorithm, we do experiments on the COREL and MSRC image datasets. Key words: tag relevance ranking; tag saliency ranking; adaptive tag ranking 1 Introduction Recent years, with the popularity of photographic electronic products such as cameras, digital photos have appeared online (e.g. Flickr) at an explosive rate. At the same time, retrieving images from enormous collections of digital photos has become an important research topic and practical problem [1]. Tag ranking has become a hot research topic due to its importance for image analysis and retrieval. Existing annotation methods about tag ranking can be roughly classified into two categories: tag relevance ranking and tag saliency ranking. Tag relevance ranking, which wa