Non-uniformity of color is a consequence of using cameras with more pixels in thinner phones: CMOS sensors are inherently very sensitive to infrared radiation, so for camera applications, it is necessary to eliminate this to avoid distorted color reproduction. This requires the use of a filter (called an IR filter) to block undesirable infra-red light. Conventional filters used to block this unwanted infrared are of the interferential type. These have the advantages of
High transmission in the wanted part of the light spectrum
A sharp cutoff at the chosen wavelength
Very low transmission at longer wavelengths
This response would be theoretically almost ideal, except for one unfortunate side effect of their operating principle – the exact wavelength at which the cutoff occurs changes depending on the angle of incidence of the incoming light rays. The compact size of camera modules used in mobile phones, coupled with relatively wide angle lenses, means the lens is very close to the sensor, which thus receives light at angles that can become quite steep in the corners of the image. The result is an unpredictable variation in the color response across the field – this is made all the more variable as it depends on the exact wavelengths present in the incoming light, and hence on both picture content and scene illumination.
Other sensor-related issues compound the problem, and even tiny unit-to-unit manufacturing variations lead to quite marked variations. Moreover, this flaw is highly dependent on the type of light sources (illuminants) in the scene. It’s important to appreciate that this phenomenon of color shading is quite separate from basic lens shading that causes a color-neutral reduction of brightness away from the center of the image, which is much easier to correct for.
Other factors also come into play: the exact angle at which the IR filter is mounted with respect to the sensor is subject to a spread of manufacturing mechanical tolerances, potentially leading to off- center shading.
The problem is made even worse under lighting conditions other than pure daylight or incandescent lighting, involving light sources such as fluorescents, LED, and discharge lamps. Unlike the former, these light sources do not have a continuous spectrum, but simulate a more-or-less ‘white’ light using a number of fairly narrow-band spectral components, chosen and combined so as to give satisfactory color rendering to the human eye. In such cases, these relatively narrow band components at the red end of the spectrum may lie close to the interferential filter cutoff, and so be dramatically affected by the change caused by the angle of incidence.
All these complex interactions make it virtually impossible to model this color shading phenomenon accurately, and hence very difficult to correct. One of the current solutions is to perform per-unit calibration: adding yet another step in the manufacturing process, this is an attempt to measure the color uniformity for each individual camera module and then program it with the necessary correction factors. This measurement is typically performed for 3 to 5 illuminants. Not only is this a very costly exercise, but inevitably, this can only be feasible for a small number light sources; thus in real-life situations the results are at best only approximate – and in some cases can even make the problem worse rather than better! This is a costly exercise that still only offers a limited solution to the problem – as can be seen from the user comments about various phone models, all employing this technique.
Per-unit calibration (PUC) has several limitations:
Color theory shows that PUC cannot work in the presence of fluorescent lighting.
Because of the costs of individually recording the defects in each camera, this process is limited to rear-facing cameras and to a few high-end models.
Blue glass and hybrid filters
Hybrid filters are also infrared filters. But unlike the normal IR filter, which works on an interference principle, these are simpler transmission/absorption filters. Thus they have a much gentler cutoff, and still transmit an unacceptable amount of far infrared energy.
This unwanted IR energy would distort color reproduction in the image, and so used alone, these would not produce acceptable results. So the idea is to team up one of these absorption filters with an interferential filter in order to combine the benefits of each. A typical cutoff is shown above. What happens is that the absorption filter’s spectral response combines with the interferential filter’s to mitigate the angle of incidence issue, so the interferential filter’s cutoff variation has no (or a reduced) impact.
The absorption filter cutoff can be tuned by chemistry, to either move it further towards the infrared (so less color shading will be corrected) or further towards the visible light, in which case less light will reach the sensor, particularly at this red end of the spectrum. Unfortunately, less light reaching the sensor means more gain will be required to compensate the lack of exposure, generating more noise in the image.
So besides failing to fully correct the color shading, in some cases hybrid filters may sometimes result in increased noise in the image, and the presence of an extra optical component increases the thickness of the camera module, as well as implying additional cost.
Some phone models employ both per-unit calibration and hybrid filters together, gaining the advantages of both but also compounding their respective drawbacks – yet still not producing a reliable solution under all circumstances.