Color Spaces

Photographers know that the world is difficult to capture using a digital camera. It be wonderful if we could capture the scene as we saw it. A scene might have a huge dynamic range, the tones between dark shadows and bright highlights. Digital cameras do record the world in a form that’s quite different from how we see the world. The initial RAW data captured has to be rendered to an image; an image as we want it to look, likely encoded into an RGB color space. This RAW data can be rendered to attempt to match the scene or the image can be rendered to create a pleasing reproduction of the scene. There’s a big difference between the two and it’s important to understand the differences.

The actual scene we attempt to record may go beyond the scale of color, luminance, saturation, that our devices can record, and beyond the ability of output devices to reproduce. When rendering we attempt to reproduce the scene colorimetry, the measured color of the scene, we often end up with an image that’s not very pleasing when viewed on a display or printed.

In technical jargon, the measured scene color the camera captures is known as Scene-Referred. Since we need to view this image on something like a display or a print, it’s usually necessary to make the image appear more pleasing on the output device and to produce the desired color appearance the image. These image colors are known as Output- Referred. The need to fit the color gamut and dynamic range to output referred data is called rendering. The camera usually performs this rendering when you select a color encoding setting such as sRGB or Adobe RGB (1998). If the camera is set to capture just RAW data, the rendering becomes the job of the photographer. The photographer expresses their idea of how the scene should be reproduced on an output device such as a standard display or print. The desired color appearance of an image you are editing is dependent on the output medium. This is not the measured color of the scene itself (scene-referred). An example of how a user would handle this output-referred processing would be using a RAW converter to produce the appearance they prefer from the RAW data.

When you set your digital camera controls to capture an image into a color space, there are two parts to this process: rendering the data and then encoding the data. In creating an output-referred image, the camera or computer system has to perform the color rendering processing before it can encode the result of the processing into a color space. Because the color space is rendered to output-referred data it cannot be used to accurately represent the scene appearance, or what would be called the colorimetry (measured color) of the scene. Therefore, first the data is rendered and this rendering process is based upon how a camera manufacturer feels they will produce the most pleasing image appearance for their customers.

As such, this rending varies for different camera manufacturers, and even different models from the same manufacturer, the rendering is not standardized. Think of this rendering process as a perceptual rendering of sorts; the rendering is that which the manufacturer feels produces visually pleasing color, not generally the colorimetrically correct color. This isn’t a problem; different film types have incorporated different looks, which are selected by the photographer based on their preferences. Two color space encodings of the same scene from different camera brands should match but that is rarely the case; no more than two shots of the same scene on two different types of film. However the degree of mismatch between cameras can be more pronounced when printing because the range of scenes captured is so much greater.

The second process after rendering the data is the actual data encoding which is standardized; the rendered data is encoded into a color space. Two identical renderings of the same scene will produce identical encodings in a given color space. When you produce an image file in a given color space you aren't producing a colorimetric copy of the scene you took the picture of, you are producing an image as it would look rendered to an sRGB display or correctly previewed in an ICC aware application like Photoshop. The image file describes the picture on an sRGB display (output-referred). That display should behave as described by the specifications that define sRGB. If the display is profiled and the data being previewed has an embedded profile, the file, will preview correctly in an application like Photoshop. What is seen, and output, isn’t a colorimetric representation of the actual scene (scene-referred). This is one reason why producing "accurate" color from a digital camera can be difficult. Every user has a different definition of what they mean or want when they say accurate. This rendering and encoding process isn’t limited to the process of creating images using digital cameras.

It is important to note the difference between a color space and a color encoding. A color space specifies the color coordinates, but not what medium the image in that color space is intended for. Photographers understand that it is necessary to view prints and transparencies under some kind of defined and correct viewing specifications. If a transparency is 2 stops too dark, viewing that dark transparency on a light box that is 2 stops too bright isn’t the right way to evaluate what is an under exposed transparency. If an image we are viewing on a display system is too dark, don’t increase the luminance controls to make the image appear lighter.

We have defined references that describe how both the light box and the display should behave and how various types of imagery should be viewed. The color image encoding specifies the color space which describes the specifics of the medium on which the colors are being rendered. In other words, how images in a color space are ideally being viewed. The reference medium will specify such parameters as the white, black, and dynamic range of a printer or display as well as the environmental conditions under which the viewing of the image will take place. The sRGB color space reference medium is that of a display with a set of well-defined specifications on how this display behaves including the ambient light around this display. Color encodings provide a bit more detail about the color space based on a reference medium; how that data when output should eventually be viewed. Since digital camera systems need to take their original RAW capture data and encode that data into some RGB color space it’s also important to recognize that the color appearance of an image you are editing is dependent on the output medium. If you want to be sure that your images will be reproduced correctly, it is necessary to communicate both the color space and the reference medium and viewing conditions.

Camera settings for color space are critical when capturing TIFF or JPEG files. (Color space settings are largely irrelevant for raw files, since color space will be determined in the raw file processor.) Most professional digital cameras allow selection of the output color space for JPEG and standard TIFF, with usually two options: sRGB and Adobe RGB (1998).