An image is a two dimensional array of 2 dimensional elements called pixels. If each pixel only has one value, the image is known as a gray-scale image and no color is defined. The range of values in the image can be interpreted as shades of any color, it is customary to use shades of black or gray-scale. However, to produce the color spectrum in the digital world, several primary colors must be mixed. Therefore in a color image, each pixel has several values depending on how many primary colors were chosen. For example on the digital monitor or color digital cameras, all colors are built by mixing the three colors of Red-Green-Blue (RGB) with various proportions. However, for printing on paper, standard printers use the Cyan-Magenta-Yellow-Key (CMYK, Key=black) color space. Therefore when printing an RGB image, usually a transformation of color spaces will be necessary.
In a colored digital camera, a color image is produced by dividing the pixel’s area between three colors (filters). However in astronomy due to the intrinsic faintness of most of the targets, the collecting area of the pixel is very important for us. Hence the full area of the pixel is used and one value is stored for that pixel in the end. One color filter is used for the whole image. Thus a FITS image is inherently a gray-scale image and no color can be defined for it.
One way to represent a gray-scale image in different color spaces is to use the same proportions of the primary colors in each pixel. This is the common way most FITS image converters work: they fill all the channels with the same values. The downside is two fold:
To solve both these problems, the best way is to save the FITS image into the black channel of the CMYK color space. In the RGB color space all three channels have to be used. The JPEG standard is the only common standard that accepts CMYK color space, that is why currently only the JPEG standard is included and not the PNG standard for example.
The JPEG and EPS standards set two sizes for the number of bits in each channel: 8-bit and 12-bit. The former is by far the most common and is what is used in ConvertType. Therefore, each channel should have values between 0 to 2^8-1=255. From this we see how each pixel in a gray-scale image is one byte (8 bits) long, in an RGB image, it is 3bytes long and in CMYK it is 4bytes long. But thanks to the JPEG compression algorithms, when all the pixels of one channel have the same value, that channel is compressed to one pixel. Therefore a Grayscale image and a CMYK image that has only the K-channel filled are approximately the same file size.