为解决在模板匹配过程中,目标图像发生局部遮挡、背景变化、光照变化以及剧烈非刚性形变等情况而出现的匹配失败问题,本文提出了一种基于多特征融合的共生矩阵模板匹配算法。首先,采用多特征融合的方法提取图像信息。分别提取图像的颜色特征、深度特征、方向梯度直方图(HOG)特征,通过主成分分析(PCA)与K均值聚类的方法实现多通道多特征融合;随后,以共生矩阵作为相似性度量方法,通过统计相似性特征信息来代替直接施加距离计算;最后,计算滑动窗口中每一组像素点对的共生概率,并加权求和作为匹配得分,由此在目标图像上寻找最佳匹配区域。通过实验对比,本文算法的AUC (Area Under Curve) 得分为0.658 6,较目前最好的几种模板匹配算法DDIS-D、DDIS-C、BBS算法分别提高了:7.9%,8.1%,20.2%。采用特征融合的方法能够充分利用图像信息,有效提高匹配的准确率;共生矩阵可以捕获图像的纹理相似性,且这种度量方法仅与共生统计有关,与实际像素无关,能够在一定程度上克服复杂场景对匹配结果带来的影响。实验结果表明本文的方法匹配精度更高、鲁棒性更强。
Abstract
Accurately identifying the target object in complex scenes (such as those involving partial occlusion, cluttered backgrounds, imbalanced illumination, and nonrigid deformations) is one of the difficulties involved in the study of template matching. To overcome the problem of precise matching under these conditions, we propose a co-occurrence matrix template matching algorithm based on multi-feature fusion. First, we extracted the color feature, depth feature, and HOG feature of the image. Thereafter, PCA and K-means clustering were used to realize multi-feature fusion. This method employs the co-occurrence matrix as its similarity measurement; this is different from traditional methods that directly use distance calculations. Finally, the matching score was calculated as the weighted sum of the probability of each window. Moreover, the region with the highest score was regarded as the best matching area. The area under the curve score of our algorithm was 0.658 6, which was 7.9%, 8.1%, and 20.2% higher than those of DDIS-D, DDIS-C, and BBS, respectively. The experimental results showed that multi-feature fusion is an effective approach for exploiting image information. With the aid of the co-occurrence matrix, the method can capture the similarities in image textures. Moreover, this measurement focuses on the co-occurrence statistics alone, rather than the actual pixels; this overcomes the impact of complex scenes on matching. The proposed method also achieves higher matching accuracy and better robustness than traditional methods.
ZHOU W . Optic disk detection approach based on adaptive multi-scale template matching [J]. Information and control , 2020 , 49 ( 2 ): 154 - 162 . (in Chinese)
SUN K , ZHANG J P . Road extraction from high-resolution remote sensing imagery based on local adaptive directional template match [J]. Opt. Precision Eng. , 2015 , 23 ( 10 z): 509 - 515 . (in Chinese)
DEKEL T , ORON S , RUBINSTEIN M , et al . Best-buddies similarity for robust template matching [C]. Proceedings of the Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA : IEEE , 2015 : 2021 - 2029 .
ORON S , DEKEL T , XUE T , et al . Best-buddies similarity-robust template matching using mutual nearest neighbors [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2016 , 40 ( 8 ): 1799 - 1813 .
WANG G , SUN X L , SHANG Y , et al . A Robust Template Matching Algorithm Based on Best-Buddies Similarity [J]. Acta Optica Sinica , 2017 , 37 ( 3 ): 274 - 280 . (in Chinese)
TALMI I , MECHREZ R , ZELINK-MANOR L . Template matching with deformable diversity similarity [C]. Proceedings of the Computer Vision and Pattern Recognition (CVPR), Honolulu , Hawaii , USA : IEEE , 2017 : 1311 - 1319 .
LU R Q . Template matching with multi-scale saliency [J]. Opt. Precision Eng. , 2018 , 26 ( 11 ): 2776 - 2784 . (in Chinese)
TALKER L , MOSES Y , SHIMSHONI I . Efficient sliding window computation for NN-based template matching [C]. Proceedings of the 15th European Conference on Computer Vision(ECCV), Munich, Germany : Springer , 2018 : 409 - 424 .
HSEU H W , BHALERAO A , WILSON R . Image Matching Based on the Co-occurrence Matrix [M]. United Kingdom : University of Warwick Press , 1999 .
JEVNISEK R , AVIDAN S . Co-occurrence filter [C]. Proceedings of the Computer Vision and Pattern Recognition (CVPR), Honolulu , Hawaii , USA : IEEE , 2017 : 3816 - 3824 .
KAT R , JEVNISEK R , AVIDAN S . Matching pixels using co-occurrence statistics [C]. Proceedings of the Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, USA : IEEE , 2018 : 1751 - 1759 .
ANNA D , ITAI L , ASHUTOSH T . Co-occurrence Based Texture Synthesis [C]. Proceedings of the Computing Research Repository (CoRR) , online: IEEE , 2020 .
SUMONYAN K , ZISSERMAN A . Very deep convolutional networks for large-scale image recognition [C]. Proceedings of the International Conference on Learning Representations (ICLR) , San Diego, CA, USA , 2015 .
MA C , HUANG J B , YAG H , et al . Hierarchical convolutional features for visual tracking [C]. Proceedings of the International Conference on Computer Vision (ICCV), Santiago, Chile : IEEE , 2015 : 3074 - 3082 .
WANG J , HE X , WEI Z H , et al . Fast star identification algorithm based on multi-feature matching [J]. Opt. Precision Eng. , 2019 , 27 ( 8 ): 1870 - 1879 . (in Chinese)
HANG G J , ZHU H . Human pose estimation based on fusion of HOG and color feature [J]. Pattern Recognition and Artificial Intelligence , 2014 , 27 ( 9 ): 769 - 777 . (in Chinese)
GANESH R N . Advances in Principal Component Analysis [M]. Springer : Singapore , 2018 .
LI H R . Improved k-means Clustering Method and its Application [D]. Harbin : Northeast Agricultural University , 2014 . (in Chinese)
WU Y , LIM J , YANG M H . Online object tracking: A benchmark [C]. Proceedings of the Computer Vision and Pattern Recognition (CVPR) , Portland , OR , USA : IEEE , 2013 : 2411 - 2418 .
WU Y , LIM J , YANG M H . Hierarchical convolutional features for visual tracking [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2015 , 37 ( 9 ): 1834 - 1848 .