上海交通大学学报

上海交通大学学报 >

2017 , Vol. 51 >Issue 10: 1228 - 1234

DOI: https://doi.org/10.16183/j.cnki.jsjtu.2017.10.012

兵器工业

 基于非线性尺度空间的航拍场景分类

展开
  •  南京信息工程大学  江苏省气象探测与信息处理重点实验室， 南京 210044

网络出版日期: 2017-10-31

基金资助

 

收起

 Aerial Scene Classification Based on Nonlinear Scale Space

Expand
  •  Jiangsu Key Laboratory of Meteorological Observation and Information Processing,
    Nanjing University of Information Science and Technology, Nanjing 210044, China

Online published: 2017-10-31

Supported by

 

Fold

摘要

 针对尺度不变特征变换(Scale Invariant Feature Transform，SIFT)算法在航拍场景分类中提取特征时，易造成边界模糊和细节丢失且无法描述颜色信息的问题，结合视觉词袋模型，提出了非线性尺度空间下融合颜色特征的新型颜色风式特征检测子(ColorKAZE，CKAZE).通过KAZE构造非线性尺度空间来检测特征信息；对颜色模型(Hue，Saturation,Value，HSV)非等间隔量化获取颜色量化矩阵，进而生成CKAZE特征描述子；利用视觉词袋和空间金字塔匹配模型融合多特征.实验表明，该算法相比SIFT算法在场景分类准确率方面提高了约8%.CKAZE描述子增强了KAZE的特征描述能力，突破了SIFT算法特征描述单一、边缘细节模糊的局限性，显著提升了无人机航拍图像的分类效果.

关键词：  ; 无人机航拍图像; 场景分类; 颜色风式特征检测子; 非线性尺度空间

本文引用格式

陈苏婷，王卓，王奇 .  基于非线性尺度空间的航拍场景分类[J]. 上海交通大学学报, 2017 , 51(10) : 1228 -1234 . DOI: 10.16183/j.cnki.jsjtu.2017.10.012

Abstract

 In aerial scene classification, scale invariant feature transform (SIFT) uses linear Gaussian decomposition to extract feature points. The algorithm has many problems, such as fuzzy boundary and loss of detail. Besides, the SIFT cannot describe the color information. Combined with bagofvisualwords (BoVW) model, CKAZE (colorKAZE) descriptor which fuses color feature in nonlinear scale space is proposed to solve these problems. KAZE is used to detect the characteristic information by constructing nonlinear scale space. Color quantization matrix is calculated by noninterval quantization in the HSV (hue, saturation, value) space, and the CKAZE feature descriptor is generated by the quantization matrix. Finally, highlevel semantic features and spatial layout information are extracted and fused. Experimental results show that the average classification accuracy of the proposed algorithm, compared to the classification algorithm based on SIFT, is improved by about 8%. The proposed algorithm improves the feature description ability of KAZE, and breaks the limitation of the SIFT classification algorithm. Besides, for the unmanned aerial vehicle (UAV) scene image, the accuracy can be greatly improved.

Key words：  unmanned vehicle aerial image; scene classification; colorKAZE; nonlinear scale space

参考文献

 ［1］杨昭, 高隽, 谢昭, 等. 局部Gist特征匹配核的场景分类［J］. 中国图像图形学报, 2013, 18(3): 264270.
YANG Zhao, GAO Juan, XIE Zhao, et al. Scene categorization of local Gist feature match kernel［J］. Journal of Image and Graphics, 2013, 18(3): 264270.
［2］杨涛, 张艳宁, 张秀伟, 等. 基于场景复杂度与不变特征的航拍视频实时配准算法［J］. 电子学报, 2010, 38(5): 10691077.
YANG Tao, ZHANG Yanning, ZHANG Xiuwei, et al. Scene complexity and invariant feature based realtime aerial video registration algorithm［J］. Acta Electronica Sinica, 2010, 38(5): 10691077.
［3］CHANG E, GOH K, SYCHAY G, et al. CBSA: Contentbased soft annotation for multimodal image retrieval using Bayes point machines［J］. IEEE Transactions on Circuits and Systems for Video Technology, 2003, 13(1): 2638.
［4］SUHASINI P S, KRISHNA K S R, KRISHNA I V M. Combining SIFT and invariant color histogram in HSV space for deformation and viewpoint invariant image retrieval［C］∥IEEE International Conference on Computational Intelligence and Computing Research. Coimbatore:IEEE, 2012: 14.
［5］AICANTARILLA P F, BARTOLI A, DAAVISON A J. KAZE Features［C］∥Proceedings of European Conference on Computer Vision. Florence Italy:Spring Link, 2012: 114.
［6］LOWE D G. Distinctive image features from scaleinvariant key points［J］. International Journal of Computer Vision, 2004, 60(2): 91110.
［7］BAY H, ESS A, TUYTELAARS T, et al. Speededup robust features(SURF)［J］. Computer Vision and Image Understanding, 2008, 110(3): 346359.
［8］于永军, 徐锦法, 张梁, 等. 基于改进KAZE特征的合成口径雷达匹配算法［J］. 上海交通大学学报, 2015, 49(9): 12881292.
YU Yongjun, XU Jinfa, ZHANG Liang, et al. SAR image matching algorithm based on improved KAZE［J］. Journal of Shanghai Jiao Tong University, 2015, 49(9): 12881292.
［9］JIANG M, GUO R, ZHANG Z, et al. Parallel implementation for AOS scheme on a dualcore cluster［C］∥IEEE International Conference on Intelligent Networks and Intelligent Systems. Los Alamitos : IEEE, 2010: 362365.
［10］PERONA P, SHIOTA T, MALIK J. Anisotropic diffusion［M］. Netherlands: Springer Link,1994: 7392.
［11］WANG Jingyue, HUANG Weizhang. Image segmentation with eigenfunctions of an anisotropic diffusion operator［J］. IEEE Transaction on Image Processing, 2016, 25(5): 21552167.
［12］张梁, 徐锦法, 夏青元, 等. 无人飞行器双目视觉位姿估计算法改进与验证［J］. 上海交通大学报, 2015, 49(9): 13871393.
ZHANG Liang, XU Jinfa, XIA Qingyuan, et al. An improvement and verification of position attitude estimation algorithm based on binocular vision for unmanned aerial vehicle［J］. Journal of Shanghai Jiao Tong University, 2015, 49(9): 13871393.
［13］BROWN M, LOWE D G. Invariant features from interest point groups［C］∥British Machine Vision Conference. Cardiff, UK:  DBLP, 2002: 656665.
［14］AGRAWAL M, KONOLIGE K, BLAS M R. Censure: Center surround extremas for realtime feature detection and matching［C］∥Computer Vision ECCV 2008. Marseille, France: Springer, 2008: 102115.
［15］LAZEBNIK S, SCHMID C, PONCE J. Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories［C］∥Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE Computer Society Press, 2006: 21692178.
［16］CHAJRI Y, MAARIR A, BOUIHALENE B. A comparative study of handwritten mathematical symbols recognition［C］∥IEEE International Conference on Computer Graphics, Imaging and Visualization. Los Alamitos: IEEE, 2016: 448451.
Options
文章导航

/

〈 〉