一种特征显著性编码的极光图像分类方法  

Aurora images classification via features salient coding

作　　者：韩冰[1,2] 仇文亮[1,2] 

机构地区：[1]西安电子科技大学电子工程学院,陕西西安710071 [2]西安电子科技大学智能感知与图像理解教育部重点实验室,陕西西安710071

出　　处：《西安电子科技大学学报》2013年第6期180-186,共7页Journal of Xidian University

基　　金：国家自然科学基金资助项目(41031064;60902082);陕西省自然科学基础研究计划资助项目(2011JQ8019);海洋公益性行业科研专项资助项目(201005017);教育部留学回国人员科研启动基金资助项目;中央高校基本科研业务基金资助项目(K5051302008)

摘　　要：提出了一种基于静态极光图像分类的新方法.在研究了极光图像特殊性的基础上,提取极光图像的尺度不变特征转换(SIFT)特征,再利用模糊C均值聚类获得所有SIFT特征的聚类中心,根据显著编码将聚类中心的权值作为极光图像的最终特征,通过支撑向量机对3 200幅极光图像进行分类.实验结果表明,所提出的新方法不但能够有效地对弧状极光进行分类,而且在复杂冕状极光图像分类时,也取得了良好效果.The change of the form of the aurora reveals the atmospheric activities and the degree of the influence of the sun on the earth. The research on aurora image classification is an effective way to study the aurora phenomenon. In this paper, an image classification method for auroras is developed. According to the characteristics of aurora images, the SIFT features of aurora image are extracted. Then the clustering centers of all SIFT features are obtained by the fuzzy C-means clustering method. Further, the weights of those clustering centers are calculated by the Salient Coding method and they are regarded as the final features for final aurora classification. Finally, the support vector machine is used to classify the 3200 aurora images. Experimental results show that the proposed method has a good performance not only on arc shape aurora images but also on complicated crown shape aurora images.

关 键 词：日侧极光 尺度不变特征转换特征 模糊C均值 显著编码 图像分类 

分 类 号：TN911.73[电子电信—通信与信息系统]