基于稀疏表达的图像自动标注  

IMAGE ANNOTATION BASED ON SPARSE REPRESENTATION AND CORRELATION MIGRATION

作　　者：崔晓萍[1] 刘丽[1] 周家琪[1] 李方方[1] 

机构地区：[1]山东师范大学信息科学与工程学院,济南250014

出　　处：《山东师范大学学报（自然科学版）》2016年第3期21-27,共7页Journal of Shandong Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(61170145;61373081);教育部博士点基金资助项目(20113704110001)

摘　　要：本文提出稀疏表达的图像标注算法.通过提取图像的SIFT特征及HSV特征,建立稀疏模型,采用距离函数对图像的特征向量进行相似度量求出稀疏系数,实现特征向量的匹配,考虑到图像内容的相似性会体现到标注词上,因此对待测图像进行自动化标注.实验表明,该方法充分运用了图像数据的稀疏化以及从已知信息到待标记图像的映射,完成图像的自动标注,提高了图像特征提取的质量,简化图像处理所用的实验数据,降低计算时间复杂度.In this paper, we propose an image annotation algorithm based on sparse representation and correlation migration. The algorithm of image annotation is based on the use of sparse combination of labeled images to give the label of the image. The main idea of this paper is to extract the SIFT features and HSV features of the image and apply them to image annotation. Using the distance function to carry on the similarity measure to the image＇s feature vector, the matching of the feature vector is realized. Experiments show that the method in this paper can accomplish the similar image retrieval and automatic annotation.

关 键 词：SIFT特征 HSV特征 标签 稀疏化 自动标注 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]