Moving from linear to log question

Ryan Emanuel · October 18, 2019

When the analogue data becomes digitized into linear, I've read that there is a lot of unusable bits on the top end, but can someone explain the process of transforming from 12 bit linear to 10 bit log, is there a function that gets applied to all of the data? Is 10 bit log still 12 bits of data?

AJ Young · October 18, 2019

I believe 12 bit linear is transformed using a gamma curve (usually 2.2 or 2.4) which then creates the 10 bit log image. A linear image unevenly distributes data across the dynamic range of the image, so the brightest areas are given too much bandwidth/data/space while the darkest areas are given too little.

Your question may be better answered at the ACES forum: https://acescentral.com/

Here are some articles and other forum posts that may answer your question!

I hope this helps!

David Hessel · October 18, 2019

2 hours ago, Ryan Emanuel said:

When the analogue data becomes digitized into linear, I've read that there is a lot of unusable bits on the top end, but can someone explain the process of transforming from 12 bit linear to 10 bit log, is there a function that gets applied to all of the data? Is 10 bit log still 12 bits of data?

First if you convert 12 bit to 10 bits you end up with 10 bits of data. The bit depth determines the number of values a color channel can store so 12 bit has 4096 possilbe values and 10 bit has 1024 possible values regardless of whether it is log or not.

The logic for storing visual data in a log curve, as I understand it, is the eye doesn't respond to increasing brightness of light in a linear way. Roughly speaking an object that has 4x times more light hiting it will only appeear to be about 2x as bright. So as the visual brightness of an object increases it takes more and more change in scene illumiance to cause the same relative increase in brightness.

If you store data in a linear way you end up in a sitiation where it takes ever larger value changes in the data to create a visual differences in the image or to put it another way the brightness of the image is much more sensitive to changes when the pixels are dark and much less sensitive to change when the pixels are light. The other option is to encode the data in a logrithmic way that more closely matches the way the eye responds to light. In this way a change in data values will have a similar change in visual brightness for both light and dark pixels. So if you take a pixel and change the value by 2x visually the pixel will appear 2x brighter as well.

Encoding colors this way allows the limited number of possible values to be more efficently and evenly allocated.

Edited October 18, 2019 by David Hessel