IEEE Geoscience and Remote Sensing Magazine - September 2015 - 70

(a)

(b)

(c)

(d)

(e)

Figure 11. Recovery results of Aqua MODIS band 6 using different methods. (a) The Original. Recovered images using the following:

(b) the LF method in [1]; (c) the HMLLSF method in [35]; (d) the WCLF method in [36]; and (e) the RMEMR method in [38].
patches, the functions are different. The relationships can
be roughly expressed as:
I 6 (x) = f1 (I 1 (x), I 2 (x), I 3 (x), I4 (x), I5 ( x), I7 (x)), x ! X p
(19)

0.8
0.6
0.4
0.2
0

0.1 0.2 0.3 0.4 0.5

0.8
0.6
0.4
0.2
0

-4
1 #10

0.8
0.6
0.4
0.2

0.1 0.2 0.3 0.4 0.5

0

0

0

0.1 0.2 0.3 0.4 0.5

Normalized Frequency

Normalized Frequency

Normalized Frequency

(a)

(b)

(c)

1

#10-4
Power Spectrum

0

-4
1 #10

Power Spectrum

Power Spectrum

-4
1 #10

Power Spectrum

Power Spectrum

where X p denotes the local region of I, and f1 represents
the relation function of band 6 and the other bands. The
authors first utilized the good detectors in band 6 and the
other bands to acquire the function f1. f1 was then used to
reconstruct the missing information of the malfunctioned
detectors in band 6. Since more spectral bands are made
use of when the spectral relationships are being modeled,
the methods of Gladkova et al. [37] and Li et al. [38] obtain better recovery results than the previous methods. The
method proposed by Li et al. is called robust M-estimator
multiregression (RMEMR).
We undertook a real data experiment on an Aqua
MODIS image acquired on January 16, 2009, over North
Korea. The original and recovery result images of the proposed algorithm are shown in Fig. 11. The test images in

the experiment were of 400 by 400 pixels and a 500-m resolution. Fig. 11(a) shows the original band 6 image, with
black stripes covering most of the image. Fig. 11(b)-(e) are
the output images recovered by LF [1], HMLLSF [2], WCLF
[3], and RMEMR [4], respectively. In Fig. 11(b), there are
obvious stripes and artifacts. These results show that when
the image contains complex terrain, it is difficult to fit the
relationship between the two bands using only one band
relationship curve. In Fig. 11(c)-(e), most of the recovered
pixels match well. The three results are relatively close, but
Fig. 11(c) is a little too smooth in the details.
Fig. 12 shows the Fourier transforms of the original
and recovered data of the different algorithms. The horizontal axis represents the normalized frequency, and the
vertical axis represents the average power spectrum of
all the columns. In Fig. 12(b), the stripes are still clearly
reflected in the frequency domain. Fig. 12(c)-(e) reveal
similar smooth results.
Since the real data experiment did not show a difference
between LF, HMLLSF, WCLF, and RMEMR, simulated experiments were also conducted to objectively evaluate the

0.8
0.6
0.4
0.2
0

0

0.1 0.2 0.3 0.4 0.5

1

#10-4

0.8
0.6
0.4
0.2
0

0

0.1 0.2 0.3 0.4 0.5

Normalized Frequency

Normalized Frequency

(d)

(e)

Figure 12. (a)-(e) Spectrograms of Fig. 11(a)-(e).

70

ieee Geoscience and remote sensing magazine

september 2015
Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2015
