Sensitivity calibration of digital colour cameras for auroral imaging

Transcription

1 Sensitivity calibration of digital colour cameras for auroral imaging Fred Sigernes 1,*, Jeffrey Morgan Holmes 1, Margit Dyrland 1, Dag Arne Lorentzen 1, Trond Svenøe 2, Karsten Heia 3, Takehiko Aso 4, Sergey Chernouss 5 and Charles Sterling Deehr 6 1 The University Centre in Svalbard (UNIS), N-9171 Longyearbyen, Norway 2 The Norwegian Polar Institute, Ny-Ålesund, Norway 3 Norwegian Institute of Fisheries and Aquaculture, Tromsø, Norway 4 National Institute of Polar Research, Tokyo, Japan 5 Polar Geophysical Institute, Murmansk Region, Apatity, Russia 6 Geophysical Institute, University of Alaska, Fairbanks, USA * Corresponding author: fred@unis.no Abstract: A method to sensitivity calibrate Digital Single Lens Reflective (DSLR) cameras is outlined. A low intensity calibrated light source tunable in wavelength is described. 31 monochromatic lines from 4000 to 7000 Å with a bandpass of ~12 Å were used to find the spectral responses for the D70 and the D200 cameras manufactured by Nikon. The source radiance ranged from about 300 to 1.6k R/Å. The cameras were operated in manual mode with 4 seconds exposure time at ISO 1600, which are typical settings required for night time photography of the aurora. For the Nikon D200 camera, the blue, green and red spectral responsivities peak at 4600, 5300 and 5900 Å, respectively. The response was high for the blue colour channel with a clear cut-off at 4100 Å for the UV part of the spectrum. The red channel response indicates low sensitivity above 6600 Å. The D70 shows similar spectral responsivity, except that it peaks in the green colour channel and it is more sensitive to both UV and NIR radiation. Both cameras are capable of detecting night - and dayside auroral at 4 second exposure time. For optimal auroral imaging capability, the green and red spectral responsivities need to be shifted up by about 300 and 400 Å in wavelength Optical Society of America OCIS codes: ( ) Optical Standards and testing; ( ) Spectroscopy, visible; ( ) Calibration; ( ); Astronomical optics; ( ) Space optics. References and links 1. N. Partamies, M. Syrjäsuo, and E. Donovan, Using colour in auroral imaging, Can. J. Phys. 85, (2007). 2. J.W. Chamberlain, Physics of the Aurora and Airglow, (Academic Press, New York and London, 1961). 3. D. Baker and G. Romick, The Rayleigh: interpretation of the unit in terms of column emission rate or apparent radiance expressed in SI units, Appl. Opt. 15, (1976). 4. F. Sigernes, J.M. Holmes, M. Dyrland, D.A. Lorentzen, S. Chernous, T. Svenøe, J. Moen, and C.S. Deehr, Absolute calibration of optical devices with a small field of view, J. Opt. Technol. 74, (2007). 5. G.W. Chang, and Y.C. Chen, Spectral Responsivity Estimator for Color Vision Systems: Filter Selection and Noise Effect, Proceedings of National Sci. Council, Part A 25, (2001). 6. C.S. Deehr, G.G. Sivjee, A. Egeland, K. Henriksen, P.E. Sandholt, R. Smith, P. Sweeney, C. Duncan, and J. Gilmer, Ground-Based Observations of F Region Aurora Associated With the Magnetospheric Cusp, J. Geophys. Res. 85, (1980). 7. M.H. Rees, Physics and chemistry of the upper atmosphere, (Cambridge University Press, 1989). (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15623

2 1. Introduction The arrival of the Digital Single Lens Reflex (DSLR) camera opens up new possibilities in the field of night sky colour photography of aurora [1]. Its ability to make long exposures with high signal-to-noise levels and the variety of objectives to choose from makes the DSLR camera a cost-effective choice compared to many of the scientific grade camera systems used today. In auroral imaging, monochromatic sensors are used in combination with optical interference filters to obtain the different wavelength emission intensities. The design requires first that light from the front lens needs to be collimated onto the filters. The filters are usually mounted on a rotating wheel called a filter wheel. The wheel needs to position each filter in alignment to the optical axis of the system before any exposure can take place. Secondly, the narrowband light from the filters needs to be focused onto the detector plane. Furthermore, if a light intensifier is used, then an additional relay lens must be included to transfer the intensified image onto the detector plane. The whole system becomes complex with several optical elements that need to be aligned for maximum throughput. In addition, three successive exposures with different centered wavelength filters rotated into position are necessary to create a color composite image. The operation consumes time which will diminish the spatial resolution since auroras tend to move quickly over the sky. The DSLR camera on the other hand uses only one lens element or objective together with a Colour Filter Array (CFA), or Colour Filter Mosaic (CFM), in front of the image sensor. A common CFM is the Bayer filter, which contains a sub mosaic of two green, one blue and one red transparent filter on top of each pixel of the sensor, respectively. The colour value of an image pixel is a combination of how much light that passes through each filter in the sub mosaic. But the details on how each camera manufacturer carries out the color reconstruction or demosaicing are proprietary information. The processed colour- or even the raw format images of the cameras do not give accurate information about the intensity in absolute physical units. It is therefore necessary to develop a procedure on how to conduct an intensity calibration. This study describes a method to obtain the spectral responsivity of a DSLR camera. Results from two Nikon cameras (D70 and D200) are presented. Finally, an evaluation is given on their ability to image aurora. Fig. 1. Experimental setup in the optical laboratory at UNIS. (1) fiber bundle from Leica 150 W Tungsten Halogen lamp, (2) camera mount, (3) order sorting filter wheel, (4) HR320 monochromator, (5) exit slit holder, (6) baffle, (7) fiber bundle to FICS spectrograph, (8) Lambertian screen, (9) FICS spectrograph, (10) steel mount rod. (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15624

3 2. Experimental setup A fiber optic illuminator connected to a Czerney Turner monochromator is used as a light source in front of a white diffuse screen (Lambertian surface). Figure 1 shows the experimental setup. The fiber optical illuminator (Leica CLS 150) uses a 150 W Tungsten lamp. The fiber bundle from the lamp house is connected by a lens to the monochromator. The lens is used to match the field of view of the fiber to the monochromator (f-matching). The lamp house is located in a separate room with a hole in the wall for the fiber bundle. Fig. 2. Optical diagram. (1) lamp input fiber bundle, (2) f-matching lens, (3) order sorting filter wheel, (4) entrance slit, (5) flat mirror, (6) concave mirror (collector), (7) reflective grating, (8) focusing mirror, (9) flat mirror, (10) exit slit, (11) Lambertian diffuse re-emitting surface, (12) DSLR camera, (13) input fiber bundle to FICS spectrograph. The monochromator or the HR320 Czerney Turner spectrometer is made by Jobin Yvon. It uses two concave mirrors of focal length 320 mm. The grating has 1200 grooves / mm with an effective area of 68 x 68 mm 2. The entrance and exit slit widths are fixed at 0.5 mm, which results in a spectral band pass of ~ 12 Å. The slits are easily changed if needed. The Lambertian surface (SRT ) is located 40 cm from the exit slit. It is made of Spectralon and is produced by Labsphere, Inc. The diffuse reflectance factors of the screen are nearly constant (ρ λ = 0.98) throughout the visible and near infrared regions of the spectrum. The intensity of the screen in absolute units is monitored be a Fixed Imaging Compact Spectrograph (FICS SN 7743) made by ORIEL. It uses a concave holographic grating (230 grooves / mm) with a nominal spectral range of Å. The detector is a 16-bit dynamic range thermoelectric cooled CCD from Hamamatsu (model INSTASPEC IV). The field of view of FICS is 22 o, which f-matches the fused silica fiber bundle used as entrance optics. The bandpass is approximately 96 Å. The spectral resolving power of the instrument is moderate, but for our purposes it is enough. The instrument s main task is to locate the centre or peak intensity wavelength of the output spectral line and to measure its intensity in units of R/Å. The unit Rayleigh, R, is used by the auroral and astronomical community. The generalized definition is R 1/ 4π 10 # photons cm sec sr, (1) ( ) 1 and it applies directly to radiance [2, 3]. The spectrograph is calibrated by the method outlined by [4]. The optical diagram of the system is shown in Fig. 2. The angular position of the grating determines the central wavelength illuminating the screen. The procedure is to first position the grating at the desired start central wavelength at 4000 Å, and measure its intensity by the FICS spectrograph. Secondly, the DSLR takes an image of the screen in raw format. (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15625

4 The grating is then rotated to the next position at 4100 Å, and the procedure is repeated up to an end wavelength of 7000 Å. The screen is then illuminated by 31 spectral lines throughout the visible spectrum. 3. Spectral responsivity In order to measure intensity with a DSLR camera it is necessary to obtain the total raw counts of the image from each color channel, k, separately. The equation is given as [ CTS] ( k ) ( k ) ui = Ci ( λ) S ( λ) dλ, (2) where λ is the wavelength, C i the intensity of the source in absolute units, and S (k) is the spectral responsivity [cf. 5]. The subscript i tags the specific observation of the source. In our case i refers to the 31 different spectra from the monochromator. Eq. (2) can be written in vector form as ( k ) ˆ T ˆ ( k ) u = C S. (3) The spectral responsivity S ˆ ( k ) where the matrix C is given as i i can now be solved if we form a set of observations ( k ) ˆ ˆ ( k ) u = C S, (4) [ C C C C ] T C =. (5) ( ) Thus matrix C forms a library of 31 spectral lines. The vector u contains 31 numbers of ( ) total counts and Sˆ k has 31 unknown response values for each colour channel, k, respectively. Originally, each vector C i contains 1024 elements covering the spectral region from 4000 to Å. The dimension is reduced to 31 by sampling the vector at each center wavelength setting of the monochromator. Singular Value Decomposition (SVD) may now be applied to solve Eq. (4). ˆ k Fig. 3. Source functions. C i (λ) is the set of observations that consists of 31 spectra from the monochromator (HR320) illuminating the diffuse screen (Lambertian surface SRT ). (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15626

5 4. Calibration Results Our library of spectra from the monochromator or set of observations is shown in Fig. 3. The exposure time of the FICS spectrograph is 4 seconds. The Nikon D70 and D200 cameras use the same exposure time at high sensitivity settings (ISO 1600) without being over exposed by the monochromator lamp. The lens aperture is set to f/1.4. Note that we used the same lens (AF Nikkor 50 mm f/1.4d) on both cameras houses. The intensity rises from ~300 R / Å at 4000 Å to ~1600 R /Å at 5900 Å, and ends at ~900 R / Å at 7000 Å. This is expected since the result is a convolution between the transmission / reflection profiles of each active component, the grating efficiency and the lamp spectrum. The spectrum of the tungsten lamp should slowly increase over the wavelength region while the monochromator grating efficiency starts to decrease beyond the red part of the spectrum. The drops in intensities at 4600 Å and 6700 Å are due to dips in the grating efficiency. Fig. 4. Block diagram of data handling. (k ) α is the total raw counts of the decomposed k i colour plane from observation i. (k ) β is the corresponding total dark counts. The Nikon raw images of format.nef are read by a program called MaxDSLR by Diffraction Limited, and converted to 16 bit uncompressed FITS images. FITS stands for Flexible Image Transport System and is a standard astronomical data format. The images are decomposed into its 3 primary colour planes by the Interactive Data Language (IDL) from the company ITT Visual Information Solutions. Figure 4 shows a block diagram of the data handling process. The raw data from both cameras are shown in Fig. 5. The counts follow a general behavior with overlap between the filters and higher count rates for the green filter. The filters are designed to resemble the human eye colour response. The D200 has twice as many pixels as the D70. This is the reason for the D200 s overall higher total count rates. In addition, D200 has a sharper blue and red cut off than the D70. The palettes in Fig. 5 are the automatically processed JPEG images of the cameras for each observation. It is hard to visually see any difference between in the RGB colour values obtained by the two cameras. Note that we use total counts and not counts per pixel for the raw data images. The reason is that the screen is not completely uniformly illuminated by the monochromator. We need to apply an integrating sphere instead of a flat screen to achieve a uniform source. These spheres are expensive, but we aim to get one for future studies. Therefore as a start, the raw data are (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15627

6 treated as the total summed counts per sensor with units of 10 6 counts [MCTS]. The assumption is that both the cameras and the FICS spectrograph detect the same illuminated area of the screen. The source area is well within the field of views of all instruments. (k ) The total dark count β is subtracted from each observation. This is done by one 4 second exposure in total darkness. Correspondingly, the bias of the detector is found by setting the exposure time to shortest possible (1/8000 seconds). The bias is found to have no effect on our measurements due to the long exposure time of 4 seconds. Fig. 5. Raw data. Panel (A) shows raw counts from the Nikon D70 camera for each color channel. The palette in the upper left corner is the processed JPEG images from the camera for each observation. Panel (B) shows the corresponding results from the Nikon D200 camera. Fig. 6. The spectral responsivity. Panel (A) shows the spectral responsivity of the Nikon D70 camera for each color channel. Panel (B) shows the corresponding results from the Nikon D200 camera. The spectral responsivity may now be solved according to Eq. (4). The result is shown in Fig. 6 for both cameras. The responsivities for the cameras look quite similar in shape with peaks for the blue, green and red channels at 4600, 5300 and 5900 Å, respectively. The main difference is that the D200 sensitivity peaks in the blue instead of the green part of the spectrum. It is also clear that the D70 has higher sensitivity in the far red and violet regions of the spectrum. These extended wings favor the D70 when it comes to taking near infra-red or ultraviolet photos. On the other hand, the D200 will in general produce sharper images due to less detected chromatic aberration from the wings. The D200 spectral responsivity is squeezed in range ( Å) compared to the D70, and any out of wavelength range lens aberrations will not be detected. (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15628

7 5. Auroral imaging In order to relate the spectral responsivity of the cameras to auroral imaging it is necessary to obtain information about the spectrum C i (λ) of the aurora in absolute units (see Eq. (2)). Figure 7 shows spectral measurements from the Kjell Henriksen Observatory (KHO) on Svalbard (78 o N, 16 o E) together with the obtained spectral responsivity of the cameras. The two spectra represent night- and dayside aurora, and are displayed to visualize the main (k) emissions that are contributing to the total counts u i in each channel k detected by the cameras. Note that the subscript i now refer to night- or dayside aurora. In general, the overall strongest emissions lines of the aurora are from the green 5577 Å and the red 6300 / 6364 Å of atomic oxygen [OI]. On the night side the green emission is stronger than the red, while on the dayside the red dominates over the green. This is due to the Fig. 7. Samples of night- and dayside spectra of the aurora from the Kjell Henriksen Observatory (KHO) on Svalbard, Norway. Both spectra are obtained by the Auroral Spectrograph (ASG) owned by the National Institute of Polar Research in Japan (NIPR). The spectral range is 4172 to 7334 Å with a medium spectral resolution of ~25 Å. The exposure time is 1 second for both samples. The dayside spectrum (blue curve) is taken 18th of January 2008 at 07:29:41 UT. The night side spectrum (red curve) is from 3rd of February 2008 at 21:46:47 UT. All of the known atomic- and molecular bands of the aurora are marked as a function of wavelength according to [2]. The right y-axis shows the intensity in R/Å. The colored dash - dotted and dotted lines show the spectral responsitivities of the Nikon D200 and the D70, respectively. difference in high versus low energy of the incoming electrons that produce the emissions on the night- versus the dayside, respectively [6]. The transition or border between low and high energy of the electrons is at ~500 ev. High energy electrons produce the green colour of an auroral arc. The upper part of the arc is red, produced by low energetic electrons. In addition, the green emission appears more structured than the red, which is more diffuse in appearance. The explanation is that after the electron has collided with the oxygen atom, the lifetime of the excited state is ~1 second for the green and ~2 minutes for the red emission. The rapid movement of an auroral arc over the sky is therefore seen in the green. The red is more washed out. A typical auroral [OI] arc is located in the altitude range from 100 up to 600 km. Note the absence of the molecular band emissions in the dayside spectrum. The bulk flux of the electrons does not have enough energy to reach the molecular part of the upper (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15629

8 atmosphere (altitudes < 100 km). However on the night side, the high energy tail of the electron flux may easily reach up to several kev and thus penetrate below ~100 km [7]. The result is seen in Fig. 7. N 2 + and N 2 band emissions appear in the blue and red wings of the spectrum, respectively. These emissions are seen as fast moving regions along the lower border of the auroral arcs. Figure 8 shows a typical auroral arc with its main auroral emissions as a function of altitude. The intensities of the auroral emissions in Fig. 7 are in the same order of magnitude as the source functions from the monochromator in Fig. 3. As a consequence, both the cameras are sensitive enough to detect both night- and dayside aurora with only 4 second exposure time. Also note that the two brightest emission lines in aurora, the green [OI] 5577 Å and the red [OI] 6300 / 6364 Å, are each located in wavelength above the peak spectral responsivity of the green and red colour channels of the cameras, respectively. This is off course not an ideal situation. A shift of the spectral responsivity of about 277 Å for the green and 400 Å for the red channel up in wavelength would improve the auroral sensitivity of both cameras substantially. Fig. 8. The main colours of the aurora. The illustration is based on a photograph taken by a Fujifilm S2 Pro camera from the SVALSAT satellite station close to Longyearbyen, Norway, at 16:04:48 UT 8 th of November, The exposure time was 30 seconds and the sensitivity was set to ISO On the other hand, there are little overlaps between the respective colour channels and the auroral emissions. Or in other words, the blue and green channels of the cameras are more or less blind to the auroral red emission lines. The same applies to the auroral green emission line. Both cameras are therefore able to detect and separate the red and green emissions of the aurora. Table 1 shows the total counts in each channel of the cameras calculated according to Eq. (2), using to the measured spectra in Fig. 7 as input. The numbers in Table 1 is the percentage of the total counts of the cameras. As seen in Fig. 7, the response is low for the blue channel (<10%). For nightside aurora the green and red channels detect close to 60 and 30 % of the total counts, respectively. The D200 seems to be more suitable to image red aurora (dayside) than the D70. The lack of red and blue signal could be compensated with taking a sequence of images with increasing exposure time to obtain higher count rates. The green channel will then most likely be over exposed in this process. The composite colour image may then be (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15630

9 Table 1. The Nikon D70 and D200 total counts for each colour channel in %. The counts are calculated for both night- and dayside aurora. Dayside aurora Nightside aurora D70 D200 D70 D200 Red u (1) Green u (2) Blue u (3) composed by choosing the optimum colour channels in the sequence. The main disadvantage would be loss of temporal resolution and spatial resolution if the aurora moves quickly. One other solution to increase the red and blue sensitivity would be to remove the internal UV-IR blocker filter in front of the CCDs. The N + 2 and the N 2 emissions will then be easier to detect together with other auroral emissions such at the strong near infrared OI emission line at 8446 Å. The above outlined calibration routine must then be repeated before any conclusion can be made. 6. Concluding remarks The principal results obtained by this study can be summarized as follows. (1) A fiber optical lamp is connected to the input of a monochromator which is tunable in wavelength. The output of the monochromator illuminates a Lambertian screen with a bandpass of ~12Å. (2) The brightness of the screen is monitored by an intensity calibrated spectrograph. A library of source functions consisting of 31 monochromatic lines is obtained in the visible part of the electromagnetic spectrum. The intensities range from 300 to 1.6 R/Å. (3) A method that retrieves the spectral responsivity of DSLR cameras has been outlined based on the above measured spectral library. (4) Two DSLR cameras, the Nikon D70 and D200, have been calibrated. Both cameras have peak spectral responsivities in the blue, green and red channels at 4600, 5300 and 5900 Å, respectively. The D200 have higher blue response compared to the D70. The D200 also cuts its response to ultraviolet and near infrared wavelengths more sharply compared to the D70. (5) As candidates for auroral imaging the cameras are able to detect both night- and dayside aurora with exposure times in the order of a few seconds for both the auroral green [OI] 5577 Å and the red [OI] 6300 / 6364 Å emissions. The blue sensitivity is low due to lack of strong emission lines in the aurora from this wavelength region. (6) The D200 is found to be slightly more sensitive than the D70 to the red auroral [OI] emission at 6300 / 6364 Å. (7) A shift in the spectral responsivity to higher wavelengths would result in optimal auroral performance for both cameras. The shift is 277 Å for the green and 400 Å for the red channel. (8) For future studies we aim to evaluate other cameras and replace the Lambertian screen with an integrating sphere. The spectral responsivity will then be deduced for (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15631

10 each pixel of the image sensors. Consequently, more detailed spatial information will be achieved of the aurora and its colours. Acknowledgement This work was financially supported by The Research Council of Norway through the project named: Norwegian and Russian Upper Atmosphere Co-operation On Svalbard #178911/S30 (NORUSCA). (C) 2008 OSA 29 September 2008 / Vol. 16, No. 20 / OPTICS EXPRESS 15632