demosaic

Demosaicing, also known as color interpolation, is the process used in digital imaging to reconstruct a complete color image from the incomplete color samples captured by an image sensor equipped with a color filter array (CFA). Most consumer sensors use a Bayer CFA pattern (RGGB), in which each pixel records only one of the three primary colors. Demosaicing estimates the two missing color components at each pixel to produce a full RGB image.

The challenge is to interpolate colors while preserving edges and avoiding artifacts such as color zippering along high-contrast boundaries and false colors in areas with fine detail. The choice of algorithm affects sharpness, color fidelity, and noise performance. Basic methods include nearest-neighbor, bilinear, and bicubic interpolation, which are fast but can blur detail and create artifacts. More sophisticated approaches use gradient or edge-direction information to guide interpolation (edge-directed or gradient-based demosaicing). Some methods explicitly model sensor noise or apply demosaicing within larger image processing pipelines that also include denoising and color correction.

In recent years, learning-based approaches using neural networks have become popular, often delivering higher-quality results at the cost of compute. Demosaicing is typically performed on raw sensor data before later steps in the image signal processing pipeline, such as color correction, gamma encoding, and sharpening. While Bayer-based mosaics are common, other CFA patterns exist, and some sensor technologies (such as the Foveon X3) capture full color at each pixel and do not require a conventional demosaic step.