Mono vs color astro cameras: Bayer sampling, filters and file sizes

How a Bayer filter divides the pixels

A one-shot color (OSC) camera has a mosaic of color filters laid over its sensor, one filter per photosite. The usual layout, which the Siril documentation calls the Bayer pattern, is described in Bryce Bayer’s 1976 Kodak patent: a filter mosaic “in registration” with the sensor array, with green filters on every other photosite, red filling the gaps in alternate rows and blue in the rows between. The patent ties this to human vision. Green, to which the eye is most responsive, is sampled most often.

On a Bayer sensor the counts are fixed by that layout. The Siril documentation puts it plainly: one pixel in two is green, one in four red and one in four blue. For the 6248 × 4176 IMX571 sensor used in the ZWO ASI2600MC Pro, that is (computed):

Filter	Photosites	Share
Green	13,045,824	1/2
Red	6,522,912	1/4
Blue	6,522,912	1/4
Total	26,091,648

Not every color sensor is a Bayer sensor. The Fujifilm X-T5 in our camera list uses an X-Trans layout, which Siril notes is also sparse in red and blue; the fractions above apply to Bayer patterns only.

What each color channel actually samples

The image scale of a camera is the sky covered by one photosite. Each color channel, though, only has data at its own photosites. Red and blue photosites repeat every second pixel along both rows and columns, so their sampling grid is twice the pixel pitch. Green forms a checkerboard: every second pixel along a row, with diagonal neighbors 1.41 pixels apart. Both statements follow from the pattern geometry:

red or blue grid (″) = 2 × 206.265 × pixel size (µm) ÷ focal length (mm)

Worked example with 3.76 µm pixels, comparing the photosite scale with the red/blue grid, and the number of samples across a 2.5″ star (computed; 2.5″ is at the low end of the 2.5–3.5″ that Diffraction Limited gives as typical for non-alpine sites):

Telescope	Focal length	Photosite scale	Red/blue grid	Photosites per 2.5″ FWHM	Red/blue samples per FWHM
William Optics Cat 51	250 mm	3.10″	6.20″	0.81	0.40
Askar FRA400	400 mm	1.94″	3.88″	1.29	0.64
Celestron EdgeHD 8	2032 mm	0.38″	0.76″	6.55	3.28

At 2032 mm even the sparse red grid puts more than three samples across the star, at or above the 2–3 pixels per FWHM that common sampling guidance asks for. At 250 mm the photosites alone are already coarse, and the red and blue channels are sampled at less than half a sample per FWHM. The sampling calculator shows where your own setup falls.

Debayering does not leave the gaps empty. It interpolates each missing color from neighboring photosites, and Siril explains that the common algorithms use spatial and spectral correlation between channels to infer part of the missing resolution. The effective resolution of a color frame therefore sits somewhere between the photosite scale and the coarser channel grid; no single published factor captures it. As one data point, Frédéric Michaud’s derivation of the NPF rule budgets a blur of two photosite widths for color reconstruction and calls that factor very conservative, closer to one in practice.

What a mono camera with filters changes

A mono camera has no filter mosaic. The AAVSO photometry guide describes the workflow: every pixel is identical, and you add separate filters in the light path to define the passband. The same guide notes that photometry from the “monochrome mode” of a color camera is usually worse than working from its separate RGB channels.

The practical consequence is that every photosite works through every filter. Behind an H-alpha filter, all 26.1 million photosites of an IMX571 mono sensor record the line; on the color version it is mainly the 6.5 million red ones. Siril describes its OSC_Extract_HaOIII script as taking the H-alpha data from the red pixels of the Bayer matrix as a half-size image, which is the same arithmetic. The narrowband filters guide covers the filters themselves.

The cost is time and hardware. A color frame records all three channels at once; a mono camera needs a separate set of subs through each filter, so a night is divided between them, and a filter wheel or drawer adds to the spacing behind the telescope.

Cameras sold in both versions

Many sensors in our data come as a mono and a color camera: ZWO ASI2600MM/MC Pro (IMX571), ASI533MM/MC Pro (IMX533), ASI6200MM/MC Pro (IMX455) and ASI183MM/MC Pro; QHY268M/C, QHY533M/C and QHY600M/C; Player One Poseidon-M/C Pro, Ares-M/C Pro and Zeus 455M/C Pro; ToupTek ATR2600M/C, ATR533M/C and ATR585M/C. The IMX571 camera page lists one family side by side.

The published specs within a pair are not always identical. ZWO’s ASI6200 page gives a peak QE of 91 % for the mono model and 80 % for the color one. The ZWO ASI294MM Pro uses a different sensor (IMX492) from the ASI294MC Pro (IMX294). QHY lists 14 mm of back focus for the QHY533M and 17 mm for the QHY533C, and ToupTek lists 3856 × 2180 pixels for the ATR585C and 3840 × 2160 for the ATR585M. Compare the two listings rather than assuming a twin.

What it means for storage

At capture the two versions write the same amount of data: one 16-bit value per photosite. An IMX571 frame is 52.19 MB as a FITS file either way (computed with the FITS block rules used by the storage calculator).

Processing is where they differ. A debayered color frame has three channels, so a 32-bit floating-point copy of one IMX571 frame is 313.1 MB, against 104.4 MB for a mono 32-bit copy (computed). A mono session through four filters produces four sets of single-channel copies instead; for the same number of subs, its processing copies are a third of the size. Siril also offers to skip debayering altogether and drizzle the raw color mosaic, which its documentation describes as giving cleaner noise than debayered stacks; the drizzle guide explains the method.