[1]胡云泉,郑丽颖,王伊雪.基于热核嵌入的图像分类技术[J].应用科技,2017,44(02):59-64,83.[doi:10.11991/yykj.201605018]
 HU Yunquan,ZHENG Liying,WANG Yixue.Image classification based on heat kernel embedding[J].Applied science and technology,2017,44(02):59-64,83.[doi:10.11991/yykj.201605018]
点击复制

基于热核嵌入的图像分类技术(/HTML)
分享到:

《应用科技》[ISSN:1009-671X/CN:23-1191/U]

卷:
第44卷
期数:
2017年02期
页码:
59-64,83
栏目:
计算机技术与应用
出版日期:
2017-04-05

文章信息/Info

Title:
Image classification based on heat kernel embedding
作者:
胡云泉 郑丽颖 王伊雪
哈尔滨工程大学 计算机科学与技术学院, 黑龙江 哈尔滨 150001
Author(s):
HU Yunquan ZHENG Liying WANG Yixue
School of Computer Science and Technology, Harbin Engineering University, Harbin 150001, China
关键词:
热核特征谱特征特征点检测SVM图像分类
Keywords:
heat-kernel featurespectral featurefeature point detectionSVMimage classification
分类号:
TP391.41
DOI:
10.11991/yykj.201605018
文献标志码:
A
摘要:
由于热核具有反映图像几乎全部背景信息的性质,提出利用热核特征代替图像谱特征来反映图像特征以提高图像的分类准确率。在对图像提取热核特征前先对图像的特征点检测算法做了改进,即采用Canny-Harris与Harris-Laplace相结合算法对图像进行特征点的检测。然后利用改进算法得到的特征点建立图的谱特征(邻接矩阵特征与拉普拉斯矩阵特征)以及热核特征,最后对这些特征进行SVM分类并比较实验结果。
Abstract:
As heat kernel can reflect almost all background information of images, this paper proposed a new method that uses heat kernel feature to substitute the spectral feature of images, so as to improve accuracy of image classification. An improved feature point detector was adopted before extracting heat kernel feature. The new feature point detection method combined Canny-Harris algorithm with Harris-Laplace algorithm. Then the derived feature points were employed to establish spectral features of a graph (adjacent matrix feature and Laplace matrix feature) and the heat kernel feature. Finally, these features were classified by SVM and the performances were analyzed with the experimental results.

参考文献/References:

[1] 庞海. 热核方法在量子场论和统计物理中的应用[D]. 天津: 天津大学, 2012: 17-89.
[2] XIAO Bai, HANCOCK E R, WILSON R C. A generative model for graph matching and embedding[J]. Computer vision and image understanding, 2009, 113(7): 777-789.
[3] XIAO Bai, HANCOCK E R, WILSON R C. Geometric characterization and clustering of graphs using heat kernel embeddings[J]. Image and vision computing, 2010, 28(6): 1003-1021.
[4] XIAO Bai, HANCOCK E R. Clustering shapes using heat content invariants[C]//Proceedings of 2005 IEEE International Conference on Image Processing. Genova, Italy: IEEE, 2005: I-1169-72.
[5] 赵春晖, 赵擎天. 一种基于Harris-Laplace特征点检测的抗几何攻击数字水印算法[J]. 应用科技, 2009, 36(9): 24-28.
[6] BRONSTEIN M M, BRONSTEIN A M. Shape recognition with spectral distances[J]. IEEE transactions on pattern analysis & machine intelligence, 2010, 33(5): 1065-1071.
[7] 刘美玲, 高虎明. 基于Harris-Laplace特征点检测抗几何攻击的数字水印算法[J]. 科学技术与工程, 2014, 14(10): 242-246.
[8] 徐贤锋, 檀结庆. 一种改进的多尺度 Harris 特征点检测方法[J]. 计算机工程, 2012, 38(17): 174-177.
[9] LUO Bin, WILSON R C, HANCOCK E R. Spectral embedding of graphs[J]. Pattern recognition, 2003, 36(10): 2213-2230.
[10] 王一丹, 林伟, 延伟东, 等. 基于热核特征的SAR图像地物识别方法[J]. 系统工程与电子技术, 2015, 37(5): 1047-1054.
[11] 杨绪峰, 林伟, 延伟东, 等. 采用热核特征的 SAR 图像目标识别[J]. 红外与激光工程, 2014, 43(11): 3794-3801.
[12] CORTES C, VAPNIK V. Support-vector networks[J]. Machine learning, 1995, 40(3): 273-297.
[13] HAN B, DAVIS L S. Density-based multifeature background subtraction with support vector machine[J]. IEEE transactions on pattern analysis and machine intelligence, 2012, 34(5): 1017-1023.
[14] 李刚, 赵玉新. 基于边界特征点提取的约束Delaunay三角剖分算法[J]. 系统仿真学报, 2007, 19(16): 3734-3738.
[15] 万琳. 基于三角网格的图像表示方法研究[D]. 武汉: 华中科技大学, 2009: 7-19.

备注/Memo

备注/Memo:
收稿日期:2016-5-25。
基金项目:国家自然科学基金项目(51679058).
作者简介:胡云泉(1990-),男,硕士研究生;郑丽颖(1976-),女,教授,博士.
通讯作者:郑丽颖,E-mail:zhengliying@hrbeu.edu.cn.
更新日期/Last Update: 2017-05-09