The 'Black Sun' Effect in CMOS Sensors

   Image sensors used in today's video and digital still cameras have come a long way from the grainy, noisy low-resolution devices of 15 years ago.

The improvement in image quality has been especially dramatic in CMOS sensors. First generation CMOS sensors (circa 1990), were arrays of

256 X 256 photodiodes that required a good imagination to see an image hiding among the fixed pattern noise (FPN) , dark current and high read noise.

In 2008, it's commonplace to see CMOS sensors with 5Mpixel resolution or higher, and  FPN and read noise have all but been eliminated by a

combination of vastly improved pixel design and device fabrication processes. However, with all the improvements there are still lingering image artifacts

that remain in many devices. One of these artifacts is what I call the 'black sun' effect or pixel inversion.

   This occurs when a camera is pointed at a scene with a very bright, very concentrated source of light like the disk of the sun. (Note - I don't recommend

doing this with any digital video or still camera for any length of time because imaging the sun's disk on a sensor with a camera lens may cause

irreparable damage to the sensor and or camera optics!) Under these conditions, we would expect the resulting image to look uniformly saturated across

the image of the sun's disk or whatever bright object is in the scene. However, some CMOS sensors actually show these super-bright areas as DARK, rather than saturated, as shown below.

'Black Sun' effect in an Omnivision OV5610 image sensor

 

   The picture above shows the output end of an optical fiber bundle placed in front of a digital camera resolution chart. The input end

of the fiber bundle is connected to a 150W tungsten light bulb. The measured illumination from the fiber bundle is 20000 lux. The image sensor

used is an Omnivision 5610   5Mpixel CMOS sensor, operating in full resolution mode at 4 frames per sec. It is clearly evident from the picture

that the central core of the fiber bundle appears black compared to the bright saturated halo around the fiber bundle. How can this happen?

 

 

Explanation-

  

   The simplest explanation for this effect uses the 'photo-electron bucket' model of  image sensors.

 

 

 

A single pixel of an image sensor array can be thought of as an electrical charge 'bucket' in which photons (squiggly lines) coming in from

a scene are converted to photoelectrons (small 'e's). When a pixel is read out, the level of photoelectrons is compared to a reference level

of electrical charge around the pixel. In the case of imaging a very bright object, the number of photons hitting the pixel during a given time

period (the exposure time), is far greater than the storage capacity of the photoelectron bucket (See diagram below).

 

 

 

When this happens, the photoelectrons converted by the pixel start spilling out into the area surrounding the pixel, temporarily raising the reference

level of charge surrounding the pixel. As a result, the net signal that is read out from the pixel actually decreases because the signal level

cannot exceed the full (saturated) level of the pixel well and the reference level is raised by the overflow photoelectrons. If the incident photon

flux is high enough, the signal and reference levels are the same and the net output signal is zero -  producing a black pixel.

 

 

Solution -

 

   If your camera happens to be using an image sensor with this behavior, it is difficult to completely eliminate this effect in post-processing.

One can think of several types of software fixes that can set up criteria to test for the onset of pixel inversion and correct for them by substituting

saturation values for actual pixel values. The danger with this approach is that there may be legitimate scenes in which a dark core is surrounded

by saturated pixels  and this approach would incorrectly process them.

   Fortunately, some CMOS designers/manufacturers have recently devised pixel architectures and processes that appear to have eliminated this effect.

An example of this is shown below with a Micron MT9P401  5Mpixel sensor using the same scene and capture conditions as in the Omnivision

image above.

 

No 'Black Sun' effect in a Micron MT9P401 sensor

 

 

Conclusion -

 

   Not all image sensors are equal - if you can't afford image artifacts like the 'black sun ' effect, evaluate the sensor performance carefully

before designing it into a video/ still camera system.