demosaicers

Demosaicers are algorithms or hardware blocks in digital imaging systems that reconstruct a full-color image from samples captured through a color filter array (CFA). A sensor typically records only one color component per pixel due to color filter mosaic, most commonly the Bayer RGGB pattern. Demosaicing interpolates the missing two color components at each pixel to produce a complete red, green, and blue value per pixel.

Common approaches include simple spatial interpolation (nearest neighbor, bilinear, bicubic) as well as more sophisticated methods designed to reduce artifacts by considering local gradients and edges. Edge-directed and gradient-based algorithms (e.g., Adaptive Homogeneity-Directed, Malvar et al.) attempt to follow edges to avoid color artifacts. Some demosaicers operate in the frequency domain or use prediction-error corrections.

In recent years, learning-based demosaicers using neural networks have demonstrated high quality, but require greater computational resources and training data. Many modern image signal processors integrate demosaicing with denoising, color correction, and white balance to streamline the pipeline.

Artifacts and challenges include zippering along edges, color moiré, and false colors around saturated areas, particularly with high-frequency textures or in low light. The choice of CFA pattern influences algorithm design; Bayer remains dominant, but other patterns such as Fujifilm's X-Trans present different demosaicing requirements.

Applications of demosaicing span embedded camera pipelines, smartphone image processing, and raw image processing software. The overarching goal is to balance processing speed, memory usage, and reconstruction quality while producing accurate color and detail.