Remote Sensing Image Change Detection Based on Low-Rank Representation - Advances in Image and Graphics Technologies - page 336

Remote Sensing Image Change Detection Based
on Low-Rank Representation *

Yan Cheng 1,2 , Zhiguo Jiang 1,2 , Jun Shi 1,2 , Haopeng Zhang 1,2 , and Gang Meng 3

1 Image Processing Center, School of Astronautics, Beihang University, Beijing, China
2 Beijing Key Laboratory of Digital Media, Beijing, China
{cy36152112,buaazhp}@126.com, [e-mail],
[e-mail]
3 Beijing Institute of Remote Sensing Information, Beijing, China
[e-mail]

Abstract. In this paper we propose an unsupervised approach based on low-
rank representation (LRR) for change detection in remote sensing images. Giv-
en a pair of remote sensing images obtained from the same area but in different
time, the subtraction and logarithm ratio operators are firstly applied to obtain
two difference images. Meanwhile the sparse part generated by LRR is also
employed for acquiring another difference image, which can detect the change
information. Afterwards, LRR is used again to obtain the low-rank part of these
three difference images which can reflect the common characteristics. Finally
k -means is performed on the low-rank part and thus the final result of change
detection can be gained. Experimental results show the effectiveness and feasi-
bility of the proposed method.

Keywords: Change detection, Remote sensing, Low-rank representation, K-means.

1 Introduction

Change detection plays a crucial role in the analysis and understanding of multi-
temporal remote sensing images, with wide applications in both civil and military
domains, such as agricultural survey [1], forest monitoring [2], natural disaster moni-
toring [3], urban change analysis [4], etc.
Change detection is the process of identifying differences in the state of an object
or phenomenon by observing it at different times [5]. This problem has been given
more attention in the past decades. Change detection algorithms are broadly divided
into two categories: the supervised and unsupervised approaches [1, 5]. The super-
vised approaches usually need multi-temporal ground truth for training. However,
such ground truth is difficult to obtain in practical applications. The unsupervised
approaches obtains the comparison results from the remote sensing images directly,
such as image differencing, change vector analysis (CVA) [6], principal component

* The project is supported by National Natural Science Foundation of China (61074029) and
Natural Science Foundation of Liaoning Province (20102014)