Image Algorithmics


Contrary to popular belief, Hexa-deca Bayer, a Bayer pattern with 4×4 pixels of each color, is easier to demosaic and can actually produce higher quality full-resolution images than Quad Bayer.

Analysis of Hexa-deca and Quad Bayer

The problem with Quad Bayer can be seen in its spectrum; it has no chroma signals that are sufficiently removed from its luminance signal to enable their clean separation.

In contrast, the Hexadeca Bayer spectrum shows several high-frequency chroma carriers that are sufficiently removed from luminance. However, this pattern suffers from the complication of having several chrominance carriers that are close to each other and their spectra overlapping. Fortunately, as one of our older paper shows, spectral overlap can be disregarded except under certain degenerate conditions. A simple necessary, though not sufficient, condition for good recovery of chroma signals is for at least one copy of two linearly independent chroma signals to be sufficiently separated from each other and from luminance.

Our Hexa-deca Processor

Although high-quality Hexa-Deca Bayer demosaicking is achievable, it poses several challenges that require solving multiple problems. Our Hexa-Deca Bayer processor overcomes these challenges by incorporating advanced techniques from our RGBW ISP, which is designed to handle complex CFA patterns and the interpolation of missing colors over long distances. By addressing these issues, our processor is capable of reconstructing high-quality images.

The resulting algorithm is efficient and a version of it has been implemented in silicon.

Scroll to Top