By Stan Moore
Image intensification permits very short sub-exposures of deep space objects opens new ways to create high resolution images with high dynamic range.
Atmospheric seeing continually shifts and shapes images. Very short exposures allow for fine reconstruction and editing of the ever changing image. This technique is well known and commonly used for lunar and planetary imaging, where the object is sufficiently bright to be detected by a digital camera in a short exposure.
Very short exposures of dim objects are not feasible with traditional digital cameras, due to the effects of camera noise. Conventional digital cameras produce readout noise on a pixel basis that is present in each and every frame. Information from a pixel that detects very few photons is obliterated by the readout noise and so it is necessary to expose each frame long enough to accumulate a signal strong enough to overpower the readout noise. Overly short exposures are swamped by noise.
Image intensification effectively eliminates readout noise, thus permitting very short sub-exposures. The intensifier itself does not produce any readout noise because there is no readout process. The digital detector within the intensified camera does produce camera noises, but those noises are rendered entirely insignificant by the thousands of secondary photons produced from every incoming photon. Using a properly configured intensified camera, it is possible to detect and identify individual photons as they arrive using a process called “photon counting.”
Benoit Schillings and I have assembled our own photon intensified cameras, which we call “ZeroCam” (zero read noise). My ZeroCam basically consists of a 1.25” eyepiece adapter attached to a C-mount adapter on the front-end of an ITT Gen-3 monocular core, with a relay lens assembly coupled to a Point Grey “fire wire” IEEE-1394b video camera. The video camera is controlled by a notebook computer and the video streams are captured on an external USB 500 GB hard drive.
An intensified camera consists of three basic components: image intensifier, relay lens and digital detector (e.g. digital video camera). The relay lens focuses light from the phosphor screen onto the cameras’ detector (e.g. CCD) where it forms a secondary image. The greatly amplified secondary image is digitized by the detector.
The graph below illustrates typical Signal to Noise comparison for low signal strengths. The X-axis is the number of photons gathered by the objective; the Y axis is the Signal to Noise. The red line is the maximum theoretical S/N at 100 percent QE (Quantum Efficiency is the percentage of photons that are registered by the detector) and no camera noises; the blue line is an intensified camera with 15 percent QE; the green line is for a single exposure using a sensitive CCD with 50 percent QE and 10 e- readout noise; the purple line is the same CCD camera but operating at high speed (1 frame per photon/pixel, same as an intensified camera in photon counting mode). As you can see, the intensified camera has great advantage for short exposures that gather few photons. But a single long exposure with high QE CCD produces superior S/N when the influx exceeds a certain threshold. But using that same a high QE CCD to stack high speed (very short) sub-exposures fails to produce significant S/N and is virtually useless.
· Resolution is limited by microchannel width. Thus the screen need not be over-sampled. The combination of microchannels and the fiber-optic bundle create a significant small scale pattern in a static image. The pattern disappears when image acquisition is dithered, not static.
· MTF (spatial contrast resolution) of an intensifier is related the PSF of the secondary photons. This reduces the contrast of small scale features in the image. Oversampling (high magnification) can greatly reduce that effect. Operating the camera in photon counting mode can improve MTF by abstracting each photon to a single point.
· Intensification produces stochastic noise, which is the uncertainty of the number of secondary photons created by each primary (incoming) photon. Stochastic noise is not an issue for traditional CCD because each detected photon generates a single electron. To see stochastic noise, note the brightness variations of the photons in the above image. It can be demonstrated mathematically that the primary effect of stochastic noise is equivalent to a reduction in QE. However, operating the camera in photon counting mode can largely eliminate stochastic noise because each photon’s intensity is reduced to a uniform single bit.
· Intensifiers produce dark current that resembles random photon strikes. The measured dark current for my intensifier is less than modern CCDs at the lower temperature. For most the most part, dark current has not been a limiting factor.
With sufficient spatial and temporal resolution, an intensified camera is capable of detecting individual photons. The camera operates in the photon counting regime when the photon density per frame is low enough to discriminate individual photons from the object of interest (including sky glow photons). This is achieved by high frame rates (10-30 FPS) and slow f-ratio (f/24-40). In the photon counting regime it is possible to indentify individual photons:
Special processing is applied to photon counting data streams. Each photon in each frame is reduced to a single 1 bit data point then the data points are combined to create the image. This process can produce superior resolution and MTF because the secondary photons are centroided to a single point. It can also improve dim object S/N by nearly eliminating stochastic noise. However, this process causes bright multi-photon objects to become nonlinear due to photon crowding and hence suffer from decreased S/N.
Here is a single 1 bit frame processed for photon counting (M57 100ms 10” f/32):
EMCCD works on the principle of multiplying the photon-induced electrons prior to readout within the IC. This has several potential advantages over image-intensification, including simplicity and definitive sampling (pixels). Most EMCCD have QE >= 50%, substantially greater than the ZeroCam’s modest 15% QE. EMCCDs are costly, over $20,000 and for that same money it is possible to get an image intensifier with similar QE.
EMCCD is not entirely without noises. CIC and charge transfer issues result in a bias noise per pixel per exposure that is particularly notable in very low flux conditions. This virtual EMCCD “read noise” is only a fraction of a photon but it is higher than ZeroCam. The graph below assumes EMCCD virtual read-noise = 0.2p (p = photon equivalent). Sky intensity is assumed to be 10% that of the signal and sampling FWHM = 3.33 pixels.
The X-axis is number of photons per pixel per frame. The range of 0.01 – 0.3 is in the photon counting regime.
Note that star QE is substantially different from pixel S/N for both cameras due to the effect of the star PSF over multiple pixels. Even though the ICCD has no pixel noise, star S/N is different from pixel S/N due to the effect of sky glow within the FWHM. The ICCD is slightly superior or equal to the EMCCD at very low photon flux, especially stars and small objects. The EMCCD produces superior S/N when pixel flux > 0.02 photons/pixel/frame but fails to produce superior star S/N < 0.2 p/p/f. These are very low fluxes, produced by fast imaging of dim objects at slow f-ratio.
Note that these curves and the crossovers are very sensitive to many factors, especially EMCCD effective read noise, sky glow and sampling. The curves shown are based on actual measurements with the ZeroCam and a Roper TC288 EMCCD. The dynamics have been confirmed with comparisons using the same scope and site. High-speed (10hz) limiting magnitude is very similar for each camera but brighter objects are better rendered by the EMCCD.
Thus the optimal domain for a low-QE photon ZeroCam is high speed imaging of stars (globular clusters) and narrow-band filtered nebula (to minimize sky glow). Dim broad-band extended objects such as galaxies present a struggle for low-QE.