Higher bit-depth Acquire Devices and Gamma

Imaging devices that have greater bit-depth than the 8-bit are able to bury the monitor gamma compensation into the byte values of the image before they convert the data into the common 8 bit/color format. By technical terms it is said that the added accuracy is infiltrated into the image data.

Because the inverse gamma calculation is done at higher accuracy integers, there will be less round off errors and much smoother quantization. Doing so results much better quality images (but for the uncalibrated CRT monitor only) than is possible to produce from linear image files (because the images will be in the normal 8 bit/color space at the time of publishing).  

The below graph shows the difference between the two cases:  
 

• scanning at 12 bit/color and burying the gamma 2.5 compensation in the 12 bit/color space and then converting to 8 bit/color image.

 
• scanning at 8 bit linearly and compensating for gamma 2.5 prior publishing.

 

Note1: The Levels (x-axis) is now in the 1/2.5 gamma space so that e.g. the level 28 represents the 1/256 intensity i.e. level 1 in the linear image.  
 
Note2: This histogram show changes to the level of pixels in the image. The levels that are 0 in the graph, does not mean that the pixels the level represents would turn black, instead there will not be any pixels in the image that has that level, they have been mapped to the nearest level.

The difference seems to be very notable in favor to the 12 bit scan that buries the gamma into the image data. However the are drawbacks with this method:

• Only very small adjustments can be done over such images, any larger adjustments (even linear adjustments like changing the black and/or white points) will remap the quantization in the 8-bit space will effectively strip the better round off accuracy from the file and this will add large artifacts.
   
• Images will be optimized for the purpose of displaying them on uncalibrated monitor only, they are 'locked' there in the gamma-space and can not be converted back e.g. for printing without notable degradation as shown in the next chapter. 
 

Removing Gamma Compensation That is Buried Into Image Data

When the image is to be printed accurately the monitor gamma compensation needs to be removed from the image data (at least partially).

The below graph shows the effect of removing an gamma compensation from the image data for two cases.  
 

• gamma 1/2.5 is first buried into 12 bit/color data , image is then converted into 8 bit/color and the gamma is calculated away by applying gamma 2.5.

 
• gamma 1/2.5 is first buried in into 8 bit/color data and then it is calculated away by applying gamma 2.5

Surprisingly when a 8 bit linear image is first compensated for gamma 2.5 and then the compensation is removed there will be no errors to the levels 0 ... 65 in the shadows, but when the same is done for the image that was compensated for the gamma 2.5 in the 12 bit space, there is errors. This is due to the fact that the originally 12-bit image is bit-depth compressed.  This is what the monitor does also. It removes the gamma compensation by applying the gamma function in analog domain (so it is mathematically like floating point calculation).

There after the quality of the images is about equal. Although the errors in the 12 bit image are smoother they are within one level so that is not a big concern. The gaps are pretty much equal in both of the images over the range 65...255. It appears that the originally 8 bit linear image performed somewhat better in the removal of the gamma compensation due to the exact performance in the shadows, but in reality both the cases are approximately equally bad for the image file quality since it is the midtones and highlights that mostly contribute the quality.

How to Accurately Use the Higher Bit Depth Devices

In order to fully exploit the better accuracy of 10 bit or 12 bit imaging devices the following three requirements should be fulfilled:  
 
• ability to scan linearly (in other words at gamma=1.0)
• acquire module is able to provide the higher bit-depth data
• image manipulation software supports higher bit-depth operations.

 

This is already possible with many higher quality acquire devices and Photoshop. 

The higher bit-depth linear image that has no errors from a gamma compensations can be imported into the higher bit-depth space of Photoshop and tonal range, color temperature and other enhancements can be applied over the higher bit-depth data. When these enhancement are done the image must be converted into the 8-bit/color space since Photoshop (current version 5.5) filters are not enabled in the higher bit-depth. However the the largest changes to the image (like tonal-range and color-cast corrections) are already done in the higher bit-depth so the better gradation has already been well utilized.