基于互联网搜索与反馈验证的图像自动标注

基于网页的图像自动标注存在图像与文本的关联假设问题，而基于内容的方法则存在数据规模小的问题。为此，将基于内容和基于网页的自动标注方法相结合，提出基于互联网搜索和反馈验证的图像自动标注方法。利用网页关联文本从互联网搜索结果中提取候选标注词，根据候选标注词在搜索得到图片的内容特征进行反馈验证。实验结果表明，该方法具有大规模标注能力，准确率比基于网页的图像自动标注方法提高了7.92%。

Automatic Image Annotation Based on Internet Search and Verification of Feedback

Up to nowadays, the research of Web-based automatic image annotation is mainly about the problem of the relevance assumption of the image and text, and the main problem of the content-based automatic image annotation is the limit of the database. Aiming at this problem, this paper proposes the Internet-search-based automatic image annotation with the verification of feedback, combining the content-based and the Web-based automatic image annotation. It extracts candidate labels from the search results using Web-based texts associated with the image, and then verifies the final results by using the Internet search results of the candidate labels with the content-based image features. Experimental results show that this method can annotate the large-scale database, with high accuracy, and achieves 7.92% improvement on the basis of web-based automatic image annotation.