Choosing the Calibration Tactic

Looking at a today's typical image manipulation systems, most often many image sources and many output devices or distribution channels exists. Typical image path often is as follows: 

1. Images are first acquired using digital cameras, scanners or video grabbers.  Some of the acquire devices have more or less non-linear distribution of intensity, some are quite linear. 
2. Images are then manipulated on the image manipulation system that has its' own transfer characteristic.
3. Finally the images are:
   • printed using a peripheral printer of computer.
   • offset printed in a newspaper or a magazine.
   • distributed for on-line viewing.
   • put on transparency or film.
 

All of these have different transfer characteristics that affect to tonality, colors and to the whitepoint.

The very best solution

The ICC color-management is the ultimate answer to all the trouble in color reproduction, often a generic device profile provides good result however the human vision is amazingly accurate, we are able to detect very small color errors so the best accuracy can be achieved only by accurate individual profiling and/or calibration of the imaging devices.

ICC color-management fully detach the system calibration (the monitor calibration of the computer) from the working-space where the images are manipulated.

With a system that is not ICC color-managed the system calibration specifies the working-space characteristics they are the same, therefore in the past there has been a lot of discussion about system calibration.

With ICC color-management there is no technical nor any other reason to calibrate the system to any other space but close to what the native characteristics of the monitor is, for CRT monitors that is gamma 2.5 space. This ensures that the dynamic range of the monitor is utilized the best possible way. Then a linear (gamma 1.0) working-space can be specified in the ICC color-managed editing software.

How to Achieve Accuracy without ICC Color-Management

There are several common tactics by which accuracy for desktop image manipulation systems is being achieved or attempted. These can be divided in three main categories: 

1. Average System Calibration

Here the system is calibrated to reflect some average of the various distribution channels or output devices, even if their transfer characteristics differ greatly. Obviously this is not a good choice it is just an average and all non-linear working-spaces will introduce the gamma induced errors.

One such average is the often mentioned gamma space 2.2 that is somewhere in between the gamma spaces of uncalibrated Mac and PC systems.

It is often said that if the system is set into gamma-space 1.8 it would match the printing media, particularly offset printing. This is not true at all. The compensation needed for a monitor and the Dotgain of a printing system may seem to be similar in that sense that both the compensation curves bend towards the same direction, but that is all there is similar with them. Large errors to the color or hues will be noticed when this approach is used. This belief was born when the first Apple LaserWriter was introduced, it was set by internal software into the gamma-space of uncalibrated Mac that is said to be in gamma-space 1.8 but accurately it is 1.72.

2. Calibrate the System for the Main Image Path Only

Here the main image path is determined, the transfer characteristics of the devices in concern are measured and the system is then set up accurately for this path only.  This was somewhat useful in the past when there really was only one output path. But as above all non-linear working-spaces will introduce the gamma induced errors.

Image files in such system will include the non-linear compensation. If in the future there is a need to output these files to an other distribution channel or device, it is not possible to do so without degradation to the image quality. There are only a limited number of discrete values that describe the color and intensity of each pixel in an image-file and every transformation will generate more quantization errors. Quantization errors can not be recovered, intensity levels that have been mapped to another level due to a transformation can not be separated from the original content of that level anymore. 

Also it will be difficult to calibrate the other devices when the main path is non-linear and the images have gamma compensation buried into them.

A major drawback with this scheme is the fact that image enhancement filters (actually numerical calculations performed on the light intensity that is described by the pixels values) expect that images and the manipulation system are linearly calibrated. Hence on a non-linearly set up systems these filters does not provide what the designer of the filter had in his/her mind. Instead they give lesser effect and tonal and hue errors.

3. Calibrate the System for Linear Intensity

Here too is a drawback, it requires a bit much of work in the set up phase. However this avoid the gamma induced errors so it will retain the images in their best possible condition for all the various output devices and it is the easiest setup to use.  

With linear setup no device dependent transfer function compensations are buried into the image data. Only when the images are published the required compensation is embedded to a copy of the image. 

This method is not limited to any number of distribution channels and will produce images easily, accurately and predictable for all purposes.

When the aim is towards accuracy and highest possible quality in desktop digital imaging this option provides that.  

More about output compensation, the file-gamma or image-gamma

Often it is said that an image file has a file-gamma or image-gamma when there is any type of distribution channel related compensation in the image file.  

This is misleading since only the cathode ray tube monitors (CRT monitors) have a transfer function that follows the gamma law. For example if the compensation that is applied to the image file is meant to compensate the dotgain of a printing system then the image incorporates dotgain compensation (not a gamma compensation).

Quantization error

The most common image format today is the 24 bit RGB (Red, Green, Blue) format: It has 8 bits for each of the colors. 8 bits gives 2^8=256 discrete intensity levels for each color so it yields the 256^3 = 16.7 million different hues.  

When a typical monitor gamma (2.5) is compensated by applying an inverse gamma transformation of 1/2.5 over the image data, there will only be 173 effective intensity levels left in the image. This can be calculated by taking the population of the numbers 0 to 255 that represent the original linearly distributed intensity levels, normalizing it, calculating the gamma compensation for each number, scaling back to 0 ... 255, histogramming this gamma compensated population with bin-width of 1 (one) and finally counting the bins that are non-zero. So this compensation will remove as much as 82 intensity levels from the image that initially had 256 levels. This is a considerable loss and they are lost forever, no manipulation technique will bring them back. Of course the pixels are still there but their intensity level is mapped to another intensity level as is illustrated on the right. For example the levels 232, 233 and 234 will all be mapped to the single level 246 when the compensation for gamma 2.5 monitor is applied to the image. In color such compression is equal to 3*3*3=27 different hues.

Image manipulation with such a highly reduced gradation is bound to induce error.

When images are captured and manipulated linearly the image data is not deteriorated by this quantization. 

Remembering that the most important image sources, the digital camera and scanners, are inherently linear devices it is best to have the image manipulation system also linearly calibrated. This way the image data will always be in the best possible condition for any output device or output path, it just needs to be compensated properly.