Infrared imaging
with the Sony DSC-F707/717


This is a description of the use of the Sony camera's "Night Shot" capability to capture infrared (with a wavelength around 800nm) images in full daylight. The clickable panorama above was taken in the infrared mode (with the RG715 and the ND filter combination) looking south from the Getty Center in Los Angeles. It shows the relative transparency of the atmosphere and also the very high reflectivity of the chlorophyll in the vegetation at these wavelengths.

How to do it

A number of Sony digital cameras offer a low-light-level capability which uses the CCD detector without the normally present infrared blocking filter. For nighttime use, the CCD is operated at high gain (high ASA rating) and a nearby scene can be illuminated with infrared-emitting diodes (LED) on the camera. This is called the "Nigh Shot" mode (NS).

For daytime use as an infrared imager, the following things have to be done with the DSC F-707/717 camera:

The light level must be reduced because the CCD is very sensitive in the infrared and, while the camera will operate with a slow ASA setting (100) in bright light, the NS mode forces the aperture always to be fully open (f/2-f/2.4 in the Sony DSC). The firmware also appears to limit the shutter speed to be slower than 1/60 s. Appropriate filtration can be achieved in a number of ways. Probably the simplest is to use a neutral density filter (grey) with a value of 0.9D which, in the visible, reduces the intensity by a factor of 8 (but it may be less opaque in the infrared!). In order to block the visible light and allow the infrared to pass, a deep red filter is needed - preferably one that is almost opaque to the naked eye. I have used a Schott glass RG715 which passes light redward of about 700nm. I have also tried a Schott glass UG1 ultraviolet filter which has a convenient red-leak at about 750nm: the ultraviolet light can be blocked by any photographic yellow, orange or red filter used in series with it. The red-leak in the UG1 filter is sufficiently weak that the neutral filter is not really required.

In order to block the emission from the infrared LEDs, I cut a ring from a sheet of black plastic about 1mm thick which fits inside the first 58mm filter to cover the two LED holes and also, incidentally, the laser hologram emitter. Thin plastic like this can be cut into perfect circles using a pair of dividers to repeatedly score the surface. I used an outside radius of 28.3mm and an inside radius of 18mm. If you want to try out the infrared capability without bothering with this ring, you can temporarily block the LED holes with small pieces of black tape.

The ring and filters I use can be seen here (UG + red) and here (0.9ND + RG715). The first shows a Cokin filter holder with a 58mm adapter ring together with my UG filter mounted in cardboard. There is also the red filter used to block the UV transmission. The second shows a pair of 'heliopan' filters (infrared 715 & grey 8x) that can be combined into a stack with the plastic ring between them and the camera.

What happens

This is what I think happens in the camera in this mode. It is based on experimenting with the images and does not result from any special knowledge of the Sony detectors or filters. If anybody out there knows more, please let me know.

Each colour 'element' in an RGB image from the camera is constructed from an interpolation between brightness values measured in an array of detector pixels having one of three colour filters depositied on its surface. In Sony cameras, these coloured pixels come in groups of four: 2 green, 1 red and 1 blue. Although these pixels see slightly displaced parts of the image, the interpolation process in the camera firmware computes an estimate of the R G and B brightness on a common grid of output pixels: the 'elements' referred to above. The spatial resolution of this final image is somewhat lower than could be achieved by a detector with no colour filters producing a monochrome image. This is the price you pay for the colour.

If we consider the properties of the red, green and blue filters forming the quartets of pixels on the detector, the red one will freely transmit the near-infrared up to the wavelength where the CCD detector loses its sensitivity (around 1000-1100 nm). Although the green and the blue filters do not transmit light in the band normally transmitted by the red filter, they will have 'red-leaks' in the near-infrared: this is a common characteristic of the pigments used to make such filters. This can be illustrated in a schematic filter transmission spectrum. In this diagram, the three normal colour filters can be seen in the visible spectrum from about 350-750nm: the R filter is a 'high-pass' element while the B and G filters both allow near-infrared light to pass. These red-leaks are normally blocked by the IR blocker filter represented by a dashed black line.

In the Sony NS-mode, this blocking filter is removed from the CCD (you can hear a faint click as this happens) and the CCD then responds to the infrared in the R pixels and, through the filter red-leaks, in the B and G pixels as well. The visible light can be removed entirely by using the infrared-transmitting RG715 filter. This has the remarkable result of converting the camera into a true, full-resolution near-infrared imager with all the CCD pixels contributing to a final monochrome image. Since the exposure metering and the focusing are done by processing the CCD signals, both procedures work properly in the infrared - even though the lens focus position will probably be displaced from that in the visible for the same subject distance.

The images produced by the Sony NS-mode are actually in RGB format and are scaled to have an 'apple-green' tint. The images in the R, G and B channels are similar but not identical. I assume that they are derived directly from the R, G and B CCD pixels in the same way as a colour image but I don't know exactly what the firmware does. Any residual 'colour' signal presumably results from the slightly different filter transmissions in the infrared. I normally convert the image directly to greyscale mode using one of the standard methods of conversion - the simplest being a direct averaging of the RGB channels.

This theory can be tested by imaging a solar spectrum in normal (colour), NightShot and IR (NS+RG715) modes (top to bottom). I did this by reflecting sunlight onto a white wall with a small reflection grating. The resulting spectra - I have retained the colour - show rather clearly what happens. It appears that the red channel is depressed and the blue channel boosted with respect to green in order to create the correct colour balance. The "+" signs mark the position of prominent telluric absorption bands, the strongest being the Fraunhofer "A" band - due to molecular oxygen - at 760nm which can be clearly seen.

Sample images

A gallery with some sample pictures was made using an RG780/ND filter combination. These have been processed to reduce the contrast slightly between the very bright vegetation and the dark sky.

Last update: 10 July 2005 (original from winter 2003)
Go to home