Image Annotation with Nearest Neighbor Based on Semantic Information
Most of the Nearest Neighbor(NN)-based image annotation methods do not achieve desired performances.The main reason is that much valuable information is lost when extracting visual features from image.In this paper,we propose a novel weighted NN-based method.Instead of using Euclidean distance,we learn a new distance metric with image semantic information to calculate the distance between the two images.Meanwhile,we utilize textual information of each image tagged by users to form weights of NN-based model.When introducing the semantic information,our method can minimize the semantic gap for intraclass variations and interclass similarities,and improve the annotation performance.Experiments on image annotation dataset of ImageCLEF2012 show that our method outperforms the traditional classifiers.Moreover,our method is simple,efficient,and competitive compared with state-of-the-art learning-based models.
Image annotation Nearest neighbor Distance metric learning Entropy weight
Wei Wu Guanglai Gao
Department of Computer Science,Inner Mongolia University,No.235 West College Road,Hohhot,China
国际会议
The 2015 Chinese Intelligent Automation Conference(2015中国智能自动化会议)
福州
英文
345-352
2015-05-08（万方平台首次上网日期，不代表论文的发表时间）