Abstract

We discuss the appearance of systematic spatial and spectral patterns of noise in remotely sensed images as well as the possibility of mitigating the effects of these patterns on the data. We describe the structure of two simple theoretical models that predict the appearance of patterns of noise (mainly stripe noise). Moreover, two new algorithms that have been specifically developed to mitigate the noise patterns are described. The performance of the two algorithms is assessed by use of some hyperspectral images acquired by different kinds of airborne sensor. The algorithms show an unexpected ability to reject these noise patterns.

© 2001 Optical Society of America

Full Article  |  PDF Article

References

  • View by:
  • |
  • |
  • |

  1. A. Barducci, I. Pippi, “Environmental monitoring of the Venice lagoon using MIVIS data,” in Proceedings of the International Geoscience and Remote Sensing Symposium IGARSS ’97, T. I. Stein, ed. (Institute of Electrical and Electronics Engineers, Piscataway, N.J., 1997), Vol. II, pp. 888–890.
  2. A. Barducci, I. Pippi, “The airborne VIRS for monitoring of the environment,” in Sensors, Systems, and Next-Generation Satellites, H. Fujisada, ed., Proc. SPIE3221, 437–446 (1998).
    [CrossRef]
  3. P. J. Curran, J. L. Dungan, “Estimation of signal-to-noise: a new procedure applied to AVIRIS data,” IEEE Trans. Geosci. Remote Sens. 27, 620–628 (1989).
    [CrossRef]
  4. M. D. Nelson, J. F. Johnson, T. S. Lomheim, “General noise processes in hybrid infrared focal plane arrays,” Opt. Eng. 30, 1682–1699 (1991).
    [CrossRef]
  5. R. A. Schowengerdt, Techniques for Image Processing and Classification in Remote Sensing (Academic, Orlando, Fla., 1983).
  6. M. Cantella, “Staring-sensor systems,” in Passive Electro-Optical Systems, S. B. Campana, ed., Vol. 5 of The Infrared & Electro-Optical Systems Handbook (SPIE Optical Engineering Press, Bellingham, Wash., 1993), pp. 157–207.
  7. J. Nieke, M. Solbrig, A. Neumann, “Noise contributions for imaging spectrometers,” Appl. Opt. 38, 5191–5194 (1999).
    [CrossRef]
  8. K. Watson, “Processing remote sensing images using the 2-D FFT-noise reduction and other applications,” Geophysics 58, 835–852 (1993).
    [CrossRef]
  9. Z. Wan, Y. Zhang, X. Ma, M. D. King, J. S. Myers, X. Li, “Vicarious calibration of the Moderate-Resolution Imaging Spectroradiometer Airborne Simulator thermal-infrared channels,” Appl. Opt. 38, 6294–6306 (1999).
    [CrossRef]

1999 (2)

1993 (1)

K. Watson, “Processing remote sensing images using the 2-D FFT-noise reduction and other applications,” Geophysics 58, 835–852 (1993).
[CrossRef]

1991 (1)

M. D. Nelson, J. F. Johnson, T. S. Lomheim, “General noise processes in hybrid infrared focal plane arrays,” Opt. Eng. 30, 1682–1699 (1991).
[CrossRef]

1989 (1)

P. J. Curran, J. L. Dungan, “Estimation of signal-to-noise: a new procedure applied to AVIRIS data,” IEEE Trans. Geosci. Remote Sens. 27, 620–628 (1989).
[CrossRef]

Barducci, A.

A. Barducci, I. Pippi, “The airborne VIRS for monitoring of the environment,” in Sensors, Systems, and Next-Generation Satellites, H. Fujisada, ed., Proc. SPIE3221, 437–446 (1998).
[CrossRef]

A. Barducci, I. Pippi, “Environmental monitoring of the Venice lagoon using MIVIS data,” in Proceedings of the International Geoscience and Remote Sensing Symposium IGARSS ’97, T. I. Stein, ed. (Institute of Electrical and Electronics Engineers, Piscataway, N.J., 1997), Vol. II, pp. 888–890.

Cantella, M.

M. Cantella, “Staring-sensor systems,” in Passive Electro-Optical Systems, S. B. Campana, ed., Vol. 5 of The Infrared & Electro-Optical Systems Handbook (SPIE Optical Engineering Press, Bellingham, Wash., 1993), pp. 157–207.

Curran, P. J.

P. J. Curran, J. L. Dungan, “Estimation of signal-to-noise: a new procedure applied to AVIRIS data,” IEEE Trans. Geosci. Remote Sens. 27, 620–628 (1989).
[CrossRef]

Dungan, J. L.

P. J. Curran, J. L. Dungan, “Estimation of signal-to-noise: a new procedure applied to AVIRIS data,” IEEE Trans. Geosci. Remote Sens. 27, 620–628 (1989).
[CrossRef]

Johnson, J. F.

M. D. Nelson, J. F. Johnson, T. S. Lomheim, “General noise processes in hybrid infrared focal plane arrays,” Opt. Eng. 30, 1682–1699 (1991).
[CrossRef]

King, M. D.

Li, X.

Lomheim, T. S.

M. D. Nelson, J. F. Johnson, T. S. Lomheim, “General noise processes in hybrid infrared focal plane arrays,” Opt. Eng. 30, 1682–1699 (1991).
[CrossRef]

Ma, X.

Myers, J. S.

Nelson, M. D.

M. D. Nelson, J. F. Johnson, T. S. Lomheim, “General noise processes in hybrid infrared focal plane arrays,” Opt. Eng. 30, 1682–1699 (1991).
[CrossRef]

Neumann, A.

Nieke, J.

Pippi, I.

A. Barducci, I. Pippi, “The airborne VIRS for monitoring of the environment,” in Sensors, Systems, and Next-Generation Satellites, H. Fujisada, ed., Proc. SPIE3221, 437–446 (1998).
[CrossRef]

A. Barducci, I. Pippi, “Environmental monitoring of the Venice lagoon using MIVIS data,” in Proceedings of the International Geoscience and Remote Sensing Symposium IGARSS ’97, T. I. Stein, ed. (Institute of Electrical and Electronics Engineers, Piscataway, N.J., 1997), Vol. II, pp. 888–890.

Schowengerdt, R. A.

R. A. Schowengerdt, Techniques for Image Processing and Classification in Remote Sensing (Academic, Orlando, Fla., 1983).

Solbrig, M.

Wan, Z.

Watson, K.

K. Watson, “Processing remote sensing images using the 2-D FFT-noise reduction and other applications,” Geophysics 58, 835–852 (1993).
[CrossRef]

Zhang, Y.

Appl. Opt. (2)

Geophysics (1)

K. Watson, “Processing remote sensing images using the 2-D FFT-noise reduction and other applications,” Geophysics 58, 835–852 (1993).
[CrossRef]

IEEE Trans. Geosci. Remote Sens. (1)

P. J. Curran, J. L. Dungan, “Estimation of signal-to-noise: a new procedure applied to AVIRIS data,” IEEE Trans. Geosci. Remote Sens. 27, 620–628 (1989).
[CrossRef]

Opt. Eng. (1)

M. D. Nelson, J. F. Johnson, T. S. Lomheim, “General noise processes in hybrid infrared focal plane arrays,” Opt. Eng. 30, 1682–1699 (1991).
[CrossRef]

Other (4)

R. A. Schowengerdt, Techniques for Image Processing and Classification in Remote Sensing (Academic, Orlando, Fla., 1983).

M. Cantella, “Staring-sensor systems,” in Passive Electro-Optical Systems, S. B. Campana, ed., Vol. 5 of The Infrared & Electro-Optical Systems Handbook (SPIE Optical Engineering Press, Bellingham, Wash., 1993), pp. 157–207.

A. Barducci, I. Pippi, “Environmental monitoring of the Venice lagoon using MIVIS data,” in Proceedings of the International Geoscience and Remote Sensing Symposium IGARSS ’97, T. I. Stein, ed. (Institute of Electrical and Electronics Engineers, Piscataway, N.J., 1997), Vol. II, pp. 888–890.

A. Barducci, I. Pippi, “The airborne VIRS for monitoring of the environment,” in Sensors, Systems, and Next-Generation Satellites, H. Fujisada, ed., Proc. SPIE3221, 437–446 (1998).
[CrossRef]

Cited By

OSA participates in CrossRef's Cited-By Linking service. Citing articles from OSA journals and other participating publishers are listed here.

Alert me when this article is cited.


Figures (18)

Fig. 1
Fig. 1

Layout of an imaging spectrometer that shows the origin of spatially coherent patterns of noise. The pattern depends on the wavelength (CCD row) selected: The disturbance is therefore also spectrally coherent.

Fig. 2
Fig. 2

Synthetic image computed from Eq. (6) assuming a monochromatic disturbance of 679 kHz and an average coherence time of 30 periods superimposed upon a constant texture. The scanning parameters utilized in this simulation were set to 12.5 Hz for the line frequency, 28 kHz for the pixel frequency, and 280 kHz for the pixel integration time, choices that nearly reproduce the characteristics of many popular airborne scanning instruments (e.g., the TIMS). The disturbance causes fully random noise.

Fig. 3
Fig. 3

Synthetic image computed from Eq. (6) assuming a Gaussian wavelet disturbance spectrum centered about 43 Hz with a dispersion of 2 Hz and an average coherence time of ten periods. The scanning parameters utilized in this simulation were set to 12.5 Hz for the line frequency, 28 kHz for the pixel frequency, and 280 kHz for the pixel integration time, choices that nearly reproduce the characteristics of many popular airborne scanning instruments (e.g., the TIMS). The disturbance has been superimposed upon a constant texture and causes the horizontal stripe-noise pattern that is typical of remotely sensed images.

Fig. 4
Fig. 4

Synthetic image computed from Eq. (6) assuming a Gaussian wavelet disturbance spectrum centered about 498 Hz, with a dispersion of 11 Hz and an average coherence time of 12 periods. The scanning parameters utilized in this simulation were set to 12.5 Hz for the line frequency, 28 kHz for the pixel frequency, and 280 kHz for the pixel integration time, choices that nearly reproduce the characteristics of many popular airborne scanning instruments (e.g., the TIMS). The disturbance has been superimposed upon a constant texture and causes the interesting noise pattern that is sometimes revealed in remotely sensed images.

Fig. 5
Fig. 5

VIRS images (raw data) acquired over Elba island, Italy, during a test flight in April 1995 that show the vertical stripe noise that is typical of imaging spectrometer sensors (push broom): (a) image acquired in the 6th spectral channel (central wavelength, 441.2 nm); (b) image acquired in the 12th spectral channel (central wavelength, 501.2 nm).

Fig. 6
Fig. 6

VIRS dark-signal image (row data) acquired in the sixth spectral channel (central wavelength, 491.25 nm) during an aerial survey performed in June 2000 over the Tuscany coast, Italy. The picture clearly shows the dark-signal contribution that is typical of push-broom imagers, which is not spatially stationary. We have also verified that, on average, the dark-signal spatial pattern differs significantly from the sensitivity spatial pattern of the same spectral channel. The spectral band tuning for this overflight is different from that shown in Table 2, which refers to a previous campaign executed in 1995.

Fig. 7
Fig. 7

TIMS image (raw data) acquired over Sicily, Italy, in October 1989 that shows the horizontal stripe noise that is typical of scanning sensors (whisk broom). The image was acquired in the fourth spectral band at a wavelength of ∼9.8 µm.

Fig. 8
Fig. 8

Amplitude factor modulating the disturbance spectrum versus frequency. This quantity, introduced by the diffraction over the finite pixel integration time, explains the natural decay of the disturbance at greater frequencies. This phenomenon has been observed in images gathered by various scanning systems (see text).

Fig. 9
Fig. 9

Power spectrum of one-dimensional signal g(t) (see Section 4) obtained from the time ordering of the raw data of the TIMS image portrayed in Fig. 7. The data plotted show only the initial part (lower frequencies) of the spectrum: (a) Spectrum of the signal up to ∼100 Hz; owing to the periodicity of the sampling system the spectrum shows a pulselike structure whose first harmonic is centered at 12.5 Hz (the line frequency value). (b) Zoom of the spectrum in the region of the first harmonic (up to 12 Hz), which shows most of the power of the disturbance that produces the stripe noise.

Fig. 10
Fig. 10

Picture computed from spatial reordering of the spectrum of one-dimensional signal g(t) that was computed from the TIMS image data shown in Fig. 7. The spectrum of signal g(t) was low-pass filtered to isolate the stripe-noise contribution.

Fig. 11
Fig. 11

Profiles calculated from a VIRS image acquired over the Tuscany coast during a test flight executed in November 1999. The image does not have clearance for publication. (a) Correcting profiles calculated for the first nine bands of the sensor. The profiles were vertically translated according to their wavelengths: lowest curve, the first spectral channel; topmost curve, the ninth spectral channel. (b) Smoothed profiles calculated for the first nine bands of the sensor. The profiles were vertically translated according to their wavelengths: lowest curve, the first spectral channel; topmost curve, the ninth spectral channel.

Fig. 12
Fig. 12

Correcting profiles calculated from a VIRS image acquired over the Tuscany coast during a test flight executed in November 1999. The image does not have clearance for publication. (a) Scatterplots for the correcting profiles calculated in the first and in the second spectral bands; (b) Scatterplot for the correcting profiles calculated in the first and the third spectral bands. One can see that the correcting profiles are mutually uncorrelated.

Fig. 13
Fig. 13

Smoothed profiles calculated from a VIRS image acquired over the Tuscany coast during a test flight executed in November 1999. The image does not have clearance for publication. (a) Scatterplots for the smoothed profiles calculated in the first and in the second spectral bands. (b) Scatterplots for the smoothed profiles calculated in the first and in the third spectral bands. The strong correlation between smoothed profiles computed in different spectral bands, which confirms the assumption that these profiles are mainly influenced by the scene texture, can be seen.

Fig. 14
Fig. 14

Pixel sensitivity (solid curve) and correcting profile (asterisks and dashed curve) relative to the pixel’s column index for the sixth spectral channel of the VIRS (central wavelength, 491.25 nm). The pixel sensitivity was measured in the laboratory (during May 2000) with a white standard reflector (Spectralon plate) and two commercial-grade light sources. The imaged field was not flat and showed a noticeable decrease toward the image sides. This slow trend was eliminated from the image data by high-pass spatial filtering. The correcting profile was retrieved from the algorithm discussed in Subsection 5.A. The ability of the algorithm to predict the value of the pixel sensitivity is impressive. The spectral band tuning for this measurement is different from that in Table 2, which refers to an overflight executed in 1995.

Fig. 15
Fig. 15

VIRS image (processed data) acquired over Elba island, Italy, during a test flight in 1995. The pictures were processed by the algorithm described here to mitigate the effects of the vertical stripe noise that is characteristic of imaging spectrometer sensors (push broom). The weighting function half-width utilized for this example was set to 11 pixels: (a) Corrected image acquired in the 6th spectral channel (central wavelength, 441.2 nm). (b) Corrected image acquired in the 12th spectral channel (central wavelength, 501.2 nm).

Fig. 16
Fig. 16

Images (raw data) acquired over several sites in Italy during various measurement campaigns. All images were acquired by airborne scanners: (a) MIVIS image gathered in the 100th spectral channel (central wavelength, 11.38 µm) over the Venice lagoon. (b) MIVIS image gathered in the 96th spectral channel (central wavelength, 9.61 µm) over the Venice lagoon. (c) TIMS image gathered in the first spectral channel (central wavelength, ∼8.4 µm) over northern Sicily. (d) TMS image gathered in the third spectral channel (central wavelength, ∼560 nm) over the northern Tyrrhenian Sea.

Fig. 17
Fig. 17

Images (processed data) acquired over several sites in Italy during various measurement campaigns. All images were acquired by airborne scanners and were processed to extract the noise pattern: (a) MIVIS image gathered in the 100th spectral channel (central wavelength, 11.38 µm) over the Venice Lagoon. (b) MIVIS image gathered in the 96th spectral channel (central wavelength, 9.61 µm) over the Venice Lagoon. (c) TIMS image gathered in the first spectral channel (central wavelength, ∼8.4 µm) over northern Sicily. (d) TMS image gathered in the third spectral channel (central wavelength, ∼560 nm) over the northern Tyrrhenian Sea.

Fig. 18
Fig. 18

Images (processed data) acquired over several sites in Italy during various measurement campaigns. All images were acquired by airborne scanners and were processed to restore the data from the stripe noise: (a) MIVIS image gathered in the 100th spectral channel (central wavelength, 11.38 µm) over the Venice Lagoon. (b) MIVIS image gathered in the 96th spectral channel (central wavelength, 9.61 µm) over the Venice Lagoon. (c) TIMS image gathered in the first spectral channel (central wavelength, ∼8.4 µm) over northern Sicily. (d) TMS image gathered in the third spectral channel (central wavelength, ∼560 nm) over the northern Thyrrenian Sea.

Tables (2)

Tables Icon

Table 1 Technical Characteristics of the VIRS

Tables Icon

Table 2 Configuration of VIRS Spectral Channels for the Overflight of Elba Island, Italy

Equations (12)

Equations on this page are rendered with MathJax. Learn more.

gξ, η, t=ix, y, λ * HλSξ, η+g0ξ, η, t, t=ty, η=ηλ, ξ=ξx,
gx, y, λ=ix, y, λ * HλSx, λ+g0x, y, λ,
gt=-+ px, y, τ-t1+mτsτdτ, t=tx, y,
2πflTpIFOV/a.
gt=st-+ px, y, τ-t×1+ Mfexp2πifτdfdτ,
gt=stτp+ Mfexp2πifτsinπfτpπfdf.
Δxgtst  Mfexp2πiftexp2πifTp-1×sinπfτpπfdf, Δygtst  Mfexp2πiftexp2πfTl-1×sinπfτpπfdf.
Δxg0,Δyg0forfTlfTp1randomized disturbance,Δxg=0,Δyg0forfTl1fTpspatially coherent pattern,Δxg=0,Δyg=0for1fTlfTpimage brightness scaling,
px, λ= gx, y, λdy.
sx, λ= wx-ξpξ, λdξ.
cx, λ=px, λsx, λ.
ix, y, λ=gx, y, λcx, λ.

Metrics