MBR-SIFT: A Novel Descriptor for Image
Matching
Mingzhe Su, Yan Ma, Xiangfen Zhang, Shunbao Li, Yuping Zhang
Abstract—The traditional SIFT method is capable of
extracting distinctive feature for image matching. However, it is
extremely time consuming in the SIFT matching due to
applying the Euclidean distance measure. Recently, many
binary SIFT (BSIFT) methods have been developed to improve
matching efficiency, while none of them is invariant to mirror
reflection. To address these issues, this paper presents a mirror
reflection invariant binary descriptor, named MBR-SIFT. In
addition, the Hamming distance has been employed as
similarity measure for MBR-SIFT descriptor. The method is
compared with other BSIFT methods, and is shown that
MBR-SIFT is better both in accuracy and efficiency.
Index Terms—mirror reflection, local descriptor, image
matching, binarization
I. INTRODUCTION
Local feature point has been successfully used in pattern
recognition and computer vision applications such as image
retrieval [1], object recognition [2], gesture recognition [3],
texture recognition [4], and wide baseline matching [5]. A
good descriptor should have robustness on photometric
transformations such as brightness and highlight while
being invariant to geometrical transformations including:
rotation, scaling, viewpoint, and reflection [6].
Until recently, there are numerous feature descriptors
have been proposed, in which scale invariant feature
Manuscript received Nov. 18, 2016; revised Nov. 30, 2016. This work
was supported in part by the National Natural Science Foundation of China
(No.61373004) and Shanghai Normal University Innovation Team Project
(No. A700115001005).
Mingzhe Su is with the College of Information, Mechanical and
Electrical Engineering, Shanghai Normal University, Shanghai, China
(e-mail: 1365702732@qq.com).
Yan Ma is with the College of Information, Mechanical and Electrical
Engineering, Shanghai Normal University, Shanghai, China (e-mail:
ma-yan@shnu.edu.cn).
Xiangfen Zhang is with the College of Information, Mechanical and
Electrical Engineering, Shanghai Normal University, Shanghai, China
(e-mail: xiangfen@shnu.edu.cn).
Shunbao Li is with the College of Mathematics & Science, Shanghai
Normal University, Shanghai, China (e-mail: lsb@shnu.edu.cn).
Yuping Zhang is with the College of Information, Mechanical and
Electrical Engineering, Shanghai Normal University, Shanghai, China
(e-mail: yp_zhang@shnu.edu.cn).
transform (SIFT) descriptor proposed by Lowe [7] is one of
the most successful and popular local image feature
descriptor. However, image matching in the SIFT method to
large-scale image database would be extremely
time-consuming. To solve this problem, several binary SIFT
(BSIFT) methods have been proposed. Ni [8] first proposed
a binary string approach for SIFT keypoints. His method
exploits the Hamming distance to measure the similarity of
two BSIFT vectors. Chen et al. [9] proposed to compare the
absolute difference of two adjacent values in a descriptor
with the threshold, which generated 128-bit BSIFT
descriptor string. Zhou et al. [10] compared the 128 values
of a SIFT descriptor individually with two threshold values.
And correspondingly, 256-bit BSIFT descriptor string has
been obtained.
The above mentioned BSIFT methods and their improved
algorithms mostly ignore the problem of mirror reflection,
which will result in significant increase in mismatch rate in
face of mirror image pair. Guo [11] presented a mirror
reflection invariant descriptor (MIFT) which is inspired
from SIFT. Nevertheless, the matching time of MIFT is
comparable to that of SIFT.
To address these issues, this paper presents a horizontal or
vertical mirror reflection invariant binary descriptor, named
MBR-SIFT. MBR-SIFT not only binarizes SIFT descriptor,
but also takes into consideration the problem of mirror
reflection. The Hamming distance is employed to measure
the similarity of SIFT features in the matching.
Experimental results on UKBench dataset show that the
proposed method not only ensures the matching accuracy
and speed, but also solves the issue of mirror reflection.
II. BSIFT
The SIFT algorithm extracts image features by searching
the keypoints in the image, and then calculates the
descriptors from the local region around the keypoints. As
shown in Fig. 1(a), the local region is first divided into 16
cells with 8 directions in each cell, and each direction is
Proceedings of the International MultiConference of Engineers and Computer Scientists 2017 Vol I,
IMECS 2017, March 15 - 17, 2017, Hong Kong
ISBN: 978-988-14047-3-2
ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
IMECS 2017
下载后可阅读完整内容，剩余5页未读，立即下载
weixin_38614484
- 粉丝: 0
- 资源: 874
我的内容管理
展开
最新资源
- JavaScript DOM事件处理实战示例
- 全新JDK 1.8.122版本安装包下载指南
- Python实现《点燃你温暖我》爱心代码指南
- 创新后轮驱动技术的电动三轮车介绍
- GPT系列：AI算法模型发展的终极方向？
- 3dsmax批量渲染技巧与VR5插件兼容性
- 3DsMAX破碎效果插件：打造逼真碎片动画
- 掌握最简GPT模型：Andrej Karpathy带你走进AI新时代
- 深入解析XGBOOST在回归预测中的应用
- 深度解析机器学习：原理、算法与应用
- 360智脑企业内测开启，探索人工智能新场景应用
- 3dsmax墙砖地砖插件应用与特性解析
- 微软GPT-4助力大模型指令微调与性能提升
- OpenSARUrban-1200：平衡类别数据集助力算法评估
- SQLAlchemy 1.4.39 版本特性分析与应用
- 高颜值简约个人简历模版分享
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
