Graphics Reference
In-Depth Information
missed alarms compared with other methods. But, our method has the least number of
overall error than other methods. It may be due to that our method uses LRR to fuse
three initial difference images, which can reflect the common characteristics of them.
Table 1.
Comparison of detection results on the Mexico dataset (in number of pixels)
False
alarms
Missed
alarms
Overall
Error
Methods
NCT
2824
1847
4671
DTC
3698
834
4532
UDWT
2855
1834
4689
Ours
855
3622
4477
Table 2.
Comparison of detection results on the Sardinia dataset (in number of pixels)
False
alarms
Missed
alarms
Overall
Error
Methods
NCT
3005
583
3588
DTC
3821
400
4221
UDWT
2939
370
3309
Ours
1479
886
2365
5
Conclusion
In this paper, we have proposed a novel change detection method for remote sensing
image, which is based on low-rank representation. It uses the subtraction operator,
logarithm ratio operator and LRR to obtain three different difference images, and then
combines them to extract commonness parts by LRR. Finally
k
-means is used to ac-
quire the final result of change map. By using LRR, our method not only reveals the
specific characteristics of image content, but also achieves the complementary advan-
tages of different kinds of change detection methods. Experimental results show that
our method can effectively obtain difference images and simultaneously outperform
state-of-the-art methods.
Acknowledgement.
This work was supported by the National Natural Science Foun-
dation of China (No. 61071137, 61371134), and the 973 Program of China (Project
No. 2010CB327900).
References
1.
Bruzzone, L., Serpico, S.B.: An iterative technique for the detection of land-cover transi-
tions in multitemporal remote-sensing images. IEEE Transactions on Geoscience and Re-
mote Sensing 35(4), 858-867 (1997)
2.
Hame, T., Heiler, I., San Miguel-Ayanz, J.: An unsupervised change detection and recog-
nition system for forestry. International Journal of Remote Sensing 19(6), 1079-1099
(1998)
Search WWH ::
Custom Search