基于内容的藻类图像分类检索算法-吉林大学学报.pdfVIP

第33卷摇 第5期 吉林 大学学报 ( 息科 学版) Vol.33摇 No.5 2015年9月 Journal ofJilinUniversity (Information ScienceEdition) Sept. 2015 文章编号:1671鄄5896(2015)05鄄0538鄄06 基于内容的藻类图像分类检索算法 李伟伟 (山东青年政治学院信息工程学院,济南250103) 摘要:为解决藻类图像分类检索的问题,提出了一套基于内容的方法对藻类图像进行分类检索。 针对海洋生物 图像对颜色特征不敏感的特点,采用SIFT(Scale鄄Invariant Feature Transform)算法针对藻类图像的形状特征进行 提取。 使用PCA(Principal Component Analysis)技术对特征进行降维,有效避免了维数灾难。 采用K鄄means算法 进行聚类,简单高效。 用词包对聚类结果进行打包,以便后续的识别。 最后用KNN(K鄄Nearest Neighbors)算法 进行识别检索。 实验结果与事实吻合,为藻类造成的环境污染问题的研究提供了有效的技术支持。 关键词:图像分类;图像检索;藻类图像;SIFT特征;PCA降维 中图分类号:TP391 文献标识码:A Algorithm of Content Based Algae Image Classification and Retrieval LI Weiwei (Department of Information Engineering,Shandong Youth University of Political Science,Jinan250103,China) Abstract:The water damage problem is more and more worse in recent years,so the problem of algae image classification and retrieval is extremely urgent. To solve this problem the algorithm based on content is proposed. SIFT(Scale鄄invariant feature transform) algorithm is used to extract shape feature of algae image because marinebiologicalimageiscolor insensitive. Thedimensionofthefeaturevector generatedby SIFT can reach up to 128. The higher dimensions can affect classification prediction and can lead to high computational complexity. PCA(Principal Component Analysis) technology is adopted to reduce the dimension of thefeature avoiding dimension disaster. K鄄means is employed to clustering algorithm,which is simple and effective. The result of the clustering algorithm is packaged by Bag of Words in order to subsequent identification. KNN (K鄄Nearest Neighbors) algorithmisusedtorecognition. Theexperimentalresultsarefinally consistentwiththe fa