Document Type

Book Chapter

Publication Date


Publication Source

Single-Sensor Imaging: Methods and Applications for Digital Cameras


Noise is among the worst artifacts that affect the perceptual quality of the output from a digital camera. While cost-effective and popular, single-sensor solutions to camera architectures are not adept at noise suppression. In this scheme, data are typically obtained via a spatial subsampling procedure implemented as a color filter array (CFA), a physical construction whereby each pixel location measures the intensity of the light corresponding to only a single color. Aside from undersampling, observations made under noisy conditions typically deteriorate the estimates of the full-color image in the reconstruction process commonly referred to as demosaicking or CFA interpolation in the literature. A typical CFA scheme involves the canonical color triples (i.e., red, green, blue), and the most prevalent arrangement is called Bayer pattern.

As the general trend of increased image resolution continues due to prevalence of multimedia, the importance of interpolation is de-emphasized while the concerns for computational efficiency, noise, and color fidelity play an increasingly prominent role in the decision making of a digital camera architect. For instance, the interpolation artifacts become less noticeable as the size of the pixel shrinks with respect to the image features, while the decreased dimensionality of the pixel sensors on the complementary metal oxide semiconductor (CMOS) and charge coupled device (CCD) sensors make the pixels more susceptible to noise. Photon-limited influences are also evident in low-light photography, ranging from a specialty camera for precision measurement to indoor consumer photography.

Sensor data, which can be interpreted as subsampled or incomplete image data, undergo a series of image processing procedures in order to produce a digital photograph. However, these same steps may amplify noise introduced during image acquisition. Specifically, the demosaicking step is a major source of conflict between the image processing pipeline and image sensor noise characterization because the interpolation methods give high priority to preserving the sharpness of edges and textures.

In the presence of noise, noise patterns may form false edge structures; therefore, the distortions at the output are typically correlated with the signal in a complicated manner that makes noise modelling mathematically intractable. Thus, it is natural to conceive of a rigorous tradeoff between demosaicking and image denoising.

Inclusive pages




Document Version

Published Version


Article is included in the repository by permission of Taylor and Francis Group, LLC, a division of Informa plc. Permission documentation is on file.


CRC Press

Place of Publication

Boca Raton, FL