改进的视角鲁棒KAZE特征匹配算法

针对KAZE特征匹配算法对视角变化敏感, 在大视角场景下不能实现正确匹配的问题, 提出了一种视角鲁棒的PKAZE(Perspective-KAZE)算法。该算法在原KAZE描述符的基础上, 计算特征点邻域内的二阶梯度均值, 形成新的扩展的80维描述符; 然后利用透视变换模型对待匹配影像进行多视角模拟, 在模拟影像上提取改进的KAZE描述符, 再进行特征匹配。最后, 选取5对含有最多正确匹配数量的影像上的匹配对作为初始结果, 利用随机抽样一致算法对初始结果提纯。对多组图像进行了匹配实验, 结果表明：与 KAZE、尺度不变特征变换(SIFT)和加速鲁棒特征(SURF)算法相比, 所提算法对视角变化具有更强的鲁棒性; 与透视尺度不变特征(PSIFT)和仿射尺度不变特征(ASIFT)算法相比, 本算法匹配正确率更高, 分别为PSIFT的2~10倍, ASIFT的2~7倍。提出的算法对视角变化具有很好的鲁棒性, 不仅对模拟影像的视角变化很稳健, 而且适用于真实三维复杂场景拍摄的大视角影像, 具有一定的实用价值。

Abstract

The KAZE algorithm is sensitive to image view changes, so that it is not capable of the correct match between images with large view-point differences. This paper proposes a PKAZE (Perspective-KAZE) algorithm which is robust to the viewpoint changes. Firstly average second-order gradient values in the neighborhood region of key-points extracted by ordinary KAZE descriptors were calculated to extend the original KAZE feature descriptor to be a new 80-dimension one. Then, a perspective transform model was used to warp the image pairs to be matched in a series of different perspective angles. New KAZE feature descriptors were extracted on transformed image pairs and were matched later. Finally, five simulated image pairs with the most correct match numbers that are in top five simulated image pairs were selected as initial matching results, and the Random Sample Consensus (RANSAC) algorithm was used to remove false matching pairs in initial results. The matching experiments were performed on several image groups. The experimental results show that the proposed algorithm is more robust to the viewpoint changes as compared with the common KAZE, Scale Invariant Feature Transform(SIFT) and Speed Up Robust Feature (SURF) algorithms. The correct rate of proposed algorithm is 2–10 times that of the Perspective Scale Invariant Feature Transform(PSIFT) and 2–7 times that of the Affine Scale Invariant Feature Transform(ASIFT). Furthermore, the proposed algorithm is not only robust to the viewpoint changes for simulated images, but also robust to the large view-point difference images in a real 3D complex scene. These results verify that the algorithm has a very good practical value.

参考文献

[1] 李新娥, 班皓, 沙巍, 等. 一种大视场TDICCD相机的多传感器图像配准方法[J]. 液晶与显示, 2014, 29(4): 644-648.

LI X E, BAN H, SHA W, et al.. Registration method of large field view and multi-sensor images of TDICCD cameras[J]. Chinese Journal of Liquid Crystals and Displays, 2014, 29(4): 644-648. (in Chinese)

[2] 何宾, 陶丹, 彭勃. 高实时性F-SIFT图像拼接算法[J]. 红外与激光工程, 2013, 42(S2): 440-444.

HE B, TAO D, PENG B. High real-time F-SIFT image mosaic algorithm[J]. Infrared and Laser Engineering, 2013, 42(S2): 440-444. (in Chinese)

[3] 姚国标, 邓喀中, 艾海滨, 等. 倾斜立体影像自动准稠密匹配与三维重建算法[J]. 武汉大学学报: 信息科学版, 2014, 39(7): 843-849.

YAO G B, DENG K Z, AI H B , et al.. An algorithm of automatic quasi-dense matching and three-dimensional reconstruction for oblique stereo images[J]. Geomatics and Information Science of Wuhan University, 2014, 39(7): 843-849. (in Chinese)

[4] 贾平, 徐宁, 张叶. 基于局部特征提取的目标自动识别[J]. 光学精密工程, 2013, 21(7): 1898-1905.

JIA P, XU N, ZHANG Y. Automatic target recognition based on local feature extraction[J]. Opt. Precision Eng., 2013, 21(7): 1898-1905. (in Chinese)

[5] 李欣璐, 杨进华, 张刘, 等. 用于快速星跟踪的双向递推匹配识别[J]. 光学精密工程, 2015, 23(5): 1443-1449.

LI X L, YANG J H, ZHANG L, et al.. Bidirectional selective rule out matching recognition of fast star tracking[J]. Opt. Precision Eng., 2015, 23(5): 1443-1449. (in Chinese)

[6] LOWE D. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110.

[7] 刘志文, 刘定生, 刘鹏. 应用尺度不变特征变换的多源遥感影像特征点匹配[J]. 光学精密工程, 2013, 21(8): 2146-2153.

LIU ZH W, LIU D SH, LIU P. SIFT feature matching algorithm of multi-source remote image[J]. Opt. Precision Eng., 2013, 21(8): 2146-2153. (in Chinese)

[8] 杨化超, 姚国标, 王永波. 基于 SIFT 的宽基线立体影像密集匹配[J]. 测绘学报, 2011, 40(5): 537-543.

YANG H CH, YAO G B, WANG Y B. Dense matching for wide-line stereo images based on SIFT[J]. Acta Geodaetica et Cartographica Sinica, 2011, 40(5): 537-543. (in Chinese)

[9] BAY H, ESS A, TUYTELAARS T, et al. Speeded-up robust features (SURF)[J]. Computer Vision and Image Understanding, 2008, 110(3): 346-359.

[10] 纪华, 吴元昊, 孙宏海, 等. 结合全局信息的SIFT特征匹配算法[J]. 光学精密工程, 2009, 17(2): 439-444.

JI H, WU Y H, SUN H H, et al.. SIFT feature matching algorithm with global information[J]. Opt. Precision Eng., 2009, 17(9): 439-444. (in Chinese)

[11] 曾峦, 王元钦, 谭久彬. 改进的SIFT特征提取和匹配算法[J]. 光学精密工程, 2011, 19(6): 1391-1397.

ZENG L, WANG Y Q, TAN J B. Improved algorithm for SIFT feature extraction and matching[J]. Opt. Precision Eng., 2011, 19(6): 1391-1397. (in Chinese)

[12] ALCANTARILLA P F, BARTOLI A, DAVISON A J. KAZE features [C]. Computer Vision–ECCV 2012. Springer Berlin Heidelberg, 2012: 214-227.

[13] WEICKERT J. Applications of nonlinear diffusion in image processing and computer vision [J]. Acta Math. Univ. Comenianae, 2001, 70(1): 33-50.

[14] MOREL J M, YU G. ASIFT: A new framework for fully affine invariant image comparison [J]. SIAM Journal on Imaging Sciences, 2009, 2(2): 438-469.

[15] CAI G R, JODOIN P, LI S Z, et al. Perspective-SIFT: an efficient tool for low-altitude remote sensing image registration [J]. Signal Processing, 2013, 93(11): 3088-3110.

[16] PERONA P, MALIK J. Scale-space and edge detection using anisotropic diffusion [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990, 12(7): 629-639.

[17] AGRAWAL M, KONOLIGE K, BLAS M R. CENSURE: Center surround extremas for realtime feature detection and matching [C]. Computer Vision–ECCV 2008, Springer Berlin Heidelbergs, 2008, 5305: 102-115.

[18] ROMENY B M H. Front-end Vision and Multi-Scale Image Analysis: Multi-Scale Computer Vision Theory and Applications, Written in Mathematical[M]. Springer Science & Business Media, 2003.

[19] CHEN Q, MONTESINOS P, SUN Q, et al. Adaptive total variation denoising based on difference curvature[J]. Image and Vision Computing, 2010, 28: 298-306.

[20] MIKOLAJCZYK K, SCHMID C. A performance evaluation of local descriptors[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(10): 1615-1630

耿则勋, 徐志军, 卢兰鑫, 沈忱, 曾德佳. 改进的视角鲁棒KAZE特征匹配算法[J]. 光学精密工程, 2016, 24(3): 616. GENG Ze-xun, XU Zhi-jun, LU Lan-xin, SHEN Chen, ZENG De-jia. Improved KAZE feature matching algorithm with viewpoint change robustness[J]. Optics and Precision Engineering, 2016, 24(3): 616.

改进的视角鲁棒KAZE特征匹配算法

关于本站 Cookie 的使用提示

全站搜索

改进的视角鲁棒KAZE特征匹配算法

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索