Linear vs. Non-Linear image explanation

Posted 02 April 2018 - 07:55 AM

As grsotnas said, a linear image has pixel values that are directly (linearly) proportional to the number of photons collected. A pixel that collected 100 photons will be 10 times brighter than a pixel that collected 10 photons. For terrestrial photos, linear images (well, almost linear) are what we are accustomed to seeing. This is because most of the pixels have received enough light to be lit in the middle of their range with few pixels being either very bright or very dark.

With astrophotography, on the other hand, the overwhelming majority of pixels receive very little light, but some (pixels lit by stars) will receive a whole lot. Keeping in mind that a linear 16-bit image spans 65,536 individual light levels, pixels lit by stars may be over 60,000 times brighter than those lit by DSOs. The result is that a linear image will show only the very brightest stars to the eye, with the very faint DSO images being so dim as to not stand out. 

The idea behind stretching is to multiply the pixel values such that we can see the dim ones better. However, if we just multiplied each pixel by some fixed number, the image would still be linear and more importantly, the brightest pixels would simply saturate. As an example, imagine an image where the fringes of a galaxy have a pixel brightness of 100 ADU and the stars have a brightness of 60,000 ADU. If we wanted to make the galaxy fringes brighter, we could multiply the pixel brightness by, say, 100. That would make the galaxy fringes in the 10,000 ADU range, which is good. But all of the bright pixels would be multiplied in brightness by the same amount. Since a pixel can't get brighter than 65,536, all of the brighter pixels just turn pure white. 

A better approach is to multiply the dimmer pixels by a larger number than the brighter pixels. This can be done with either levels of curves. Since different pixels get multiplied by different numbers, there is no longer a linear relationship between pixel brightness and photons received.

Tim