| Home .:. Intro .:. The Color Correction Problem .:. The New Spectrum for Accuracy .:. Applying the Spectrum |
|
The New Spectrum for Accuracy We've discussed a couple of the major underlying problems of color correction. The first problem is the failure to use a perceptual color space. The second involves the distortions that occur when fitting an image into a color range that can't fit all the colors it needs to. Let's more closely look at a solution to the first problem, specifically Master Colors' solution. We've said the L*a*b* color space is convenient for many purposes, such as color measurement, matching, and general color management. But where it falls short is its ability to describe color on an intuitive level. As such, transformations on colors in this space can produce odd results with respect to human perception. This is not the case, by definition, with a perceptual color space. The language describing each color in such a space is one that resonates with a person's basic understanding of color. So, an image that is "dragged" through this type of color space will change exactly as we would expect it to. The perceptual space used by Master Colors is the HVC space (Hue, Value, Chroma). The details of this space are discussed elsewhere (see section: The Color Chart). But to give you the basic idea, this space include 3 separate axes for the descriptive color qualities, like L*a*b*. But the qualities are closer to the terms by which we would describe colors in plain language. Hue indicates which color group in the spectrum the color exists in, such as red, yellow, green, etc. Value is very similar to 'L' in L*a*b*, describing how light or dark the color is. Chroma describes the intensity, where gray is the weakest, and the bright colors of every hue are the strongest. Thus you can see a familiar paradigm for describing color: you might say "a light, bright orange" or a "dark, pale blue". Just as there is a universal L*a*b* model, into which every conceivable color fits, there is an HVC model like this too. Any color in the visible light spectrum can be described using Hue, Value, and Chroma coordinates.  An approximate model of the 3-dimensional form of the full HVC space. Any color can be classified according to a position in this space. Note that it is more irregularly shaped than the L*a*b* space. Geometric simplicity is sacrificed for visual consistency. The process for copying an image to another device using the HVC space is very similar to the one discussed earlier, using L*a*b*. Every device has a range of color available that is a sub-space of this greater HVC space, carving out a smaller nook within it. When moving an image from one device to another, the colors of the image simply have their HVC coordinates shifted until they fit in the new range.  The color coordinates are "pushed down" into the space of the darker color range. Since we are shifting the colors through a perceptual color space, we can now be confident that the colors' properties most important to us are conserved. In this case, we are only adjusting the Value (lightness) of each color, thus the Hue and Chroma of each color will be undisturbed. Such is not necessarily the case when operating in a non-perceptual color space, resulting in distortions. There is another very important result in switching to the perceptual HVC space. Because the 3 axes of HVC space actually have meaning to basic human perception, and because the axes combine to form a smooth, evenly spaced continuum of perceptually defined color, we can treat the space in many way like physical space. Specifically, we can determine the distance between any 2 colors, like we would calculate distance between two points in real space. This color distance has a specific relevance. It characterizes, in precise numeric terms, exactly how visually different these two colors appear to be. There is another way of looking at this. When two colors are placed next to each other, they form a border between them. The strength of this border is known as the contrast between the two colors. The strength of this contrast is precisely the distance between the colors in HVC space. So, the process of determining the distance between two colors can be correctly thought of as quantifying contrast. Weak contrast signifies a small distance between colors, whereas strong contrast signifies a large distance. (A more detailed discussion of this concept can be found in section: A Color Revolution) The notion of quantifying contrast is really the crux of Master Colors ground breaking ideas. And as we will see, it can shed much light on the mysterious problems plaguing color correction. To begin with, let's go back to our example of down-shifting an image through L*a*b* space. We've already noted that, since it is not a perceptual space, the colors may skew in ways we don't expect as we shift the image. But there is another problem related to this. Since L*a*b* is not a perceptually smooth continuum of color like HVC, the distances between each color will begin to warp as we shift the image through the space. Some distances will expand, others will contract. In other words, some contrast levels will soften, while others will sharpen. This is another form of distortion an image may experience. |
|
When using the HVC color space, this problem is completely eliminated. All of the distances are preserved as the image is shifted through the space. This means that the contrast levels of the copied image appear precisely the same as they originally did. We have introduced another factor by which we gauge the accuracy of a copied image. It should be clear that an accurate image will have its color distances conserved, in addition to having all its colors closely resemble the originals. These two considerations are actually two sides of the same coin. If the copied image's colors are distorted, offline from the originals, then clearly the distances between them will be distorted. And this is a problem inherent in a non-perceptual color space.  A hypothetical example of colors copied to a different medium. The distortion of the colors themselves is obvious. But an outgrowth of this problem is the distortion in distances. The contrast between the colors has changed dramatically as the colors have distorted. Preserving color distance should also be strived for during the transition. Being able to conserve the contrast levels of a copied image is one of the great advantages of using the HVC space. Such conservation occurs naturally, as it is a built-in feature of the structure of HVC. In fact, the HVC is really the only color space in which this is possible; no other color space even allows for the consistent measurement of color contrast. But what about the situation discussed in the previous section, where the new device supports a color range that is smaller or differently shaped than the original? This situation looks very similar to our L*a*b* example when shifting an image through HVC. |
Just as the case when using the L*a*b* space, sometimes the new device will not support the same color range as the first device. Some compromise will be needed to display all of the colors.
|
Situations like this are difficult to avoid, since most devices will differ in the color ranges they're able to display. We need a strategy for handling such disparities, some form of compromise to display each color. We mentioned earlier the simplest way of doing this - by pushing each overflowing color towards the closest boundary. The results of this approach are not the most desirable though. Luckily, the power of the HVC space, and the ability to identify color distances, can show us the path towards the best approach. We've already showed that preservation of an image's color distances, or contrast levels, should be considered an integral part of whether the image was accurately copied or not. But it should be clear that if a color range cannot support all of the necessary colors, at least some of the color distances must be altered in order to display each color. So we ask, what is the best way to alter the colors and their distances such that minimum distortion is incurred? The answer lies in a simple idea, but the explanation is a bit subtle. If each contrast level is proportionally conserved, then the character of that information should be conserved. In other words, each color distance should relate to each other distance in exactly the same way after the image is copied as they did originally. Note that each distance overall may be diminished, but they will bear the exact same proportions to each other. This is the next best solution to having each distance exactly conserved. Since we know we must change the distances, we now strive to keep them proportionally identical. But how can such a feat be accomplished? How do we control each distance so precisely, keeping them related to each other? With the HVC space, this is quite easy. In fact, it happens automatically, as we "shrink" the entire network of colors, until it fits into the color range.  Each color's relative position is adjusted, such that the whole structure is "shrunken" in HVC space. This is a remarkably convenient aspect of the HVC space. Any time the colors of an image need to be "compressed" or "expanded", so to speak, to better fit a certain color range, there will be absolutely no distortions in the relative contrast levels. You might wonder why specifically preserving the color distance proportions will lead to a more accurate result than not doing so. To answer this, we must picture the role that contrast plays in an image. Contrast is the visual strength of the differences between the colors. In a sense, it helps us differentiate one color from the next. In doing so, contrast gives us most of the information about the structure and form the image. The most strong and prominent forms in the image will generally be distinguished by large contrast levels. Whereas the more subtle aspects of the image, such as texture, detail, grading and such nuance, will be the result of low contrast, or small color distances. This is very important information, largely conveyed through contrast. If contrast levels are altered, this information changes. Specifically, if contrast levels are altered non-proportionally with respect to each other, the image loses its hierarchy of form and structure. On the other hand, if they are altered proportionally, this hierarchy remains unchanged.  An example of an image being copied to a medium that cannot show intense colors. Top: The only color that is changed is the red circle. It is "pushed" into the space. Consequently, its distance to the background colors is shortened. But the background colors were unchanged, so the distance between them is the same. Thus the proportional hierarchy has been skewed. Bottom: All of the colors were moved instead. They were "shrunken" to fit in the smaller color range. Thus the distances bear the exact same proportions as they originally did. The contrast of the circle is still dominant, by the same degree. (note, the relations of this diagram may appear different on a printed page) Having understood the value of conserving the proportionality of the distances, we should note that this is not the end of the story. In most cases, it will not be sufficient to only make sure that the proportions among each color distance stay the same. When we "shrink" an image, so to speak, we are actually changing every single color in the image. We are shifting the position of each just enough, such that ever color now fits in the new range. The beautifully preserves all the relationships, and all the valuable information they embody. But at the same time, this strategy is at odds with our desire to see the colors "stay the same". When we copy an image to another medium, we have an almost instinctive expectation that the image will possess the "same" colors it had before. As we've seen, in most cases this won't be possible, since it will be copied to a different range of color, which may generally be a darker, lighter, or less intense range. But even so, we expect that after the proper adjustments are made, that the colors are still reasonably close to the positions in color space they once held. But we've shown that there is a distinct tradeoff. Holding the colors' positions will result in color distance distortion. Similarly, preserving the color distance relationships will serve to move every color away from their original positions.
Left: Colors outside the boundary are pushed inside. As few colors are moved as possible. However, the distance relationships are now skewed. Right: Every color is pushed inward proportionally until all colors fit inside. Each color has been altered in position, but the relationships are preserved exactly. As you might guess, what is probably desirable in most cases is some kind of balance between these two approaches. One where color position is minimally changed, as long as change in the proportions among contrast levels is also beneath a reasonable threshold. This is a more formalized statement of the compromise we hinted at earlier. That is to seek a balance between these two factors as the situation demands.  When faced with the problem of colors overflowing the boundaries of the color range, there are two possibilities for correction. (moving down): We can push outside colors inside the border, moving as few colors possible, but distorting the proportions. (moving right): Or we can press all the colors together relative to each other, changing each color, but conserving proportions. Ideally, there should be a point in between these two measures that accomplishes both effects to a certain degree, both keeping proportions at an acceptable level, while altering the colors only a little. This inherent tradeoff, and the many steps of compromise between the extremes, are what we mean by "the new spectrum for accuracy". These are exactly the measures by which we judge the accuracy of a copied image, whether we are aware of it or not. By formalizing what was once confined to the subtle judgment of experts, or even just common human observation, we now have a tremendous amount of control over the quality of translated images. It is just a matter of understanding the devices being used, and then identifying the best "compromise" on this scale. No longer is it needed to waste endless hours obsessing over the quality of the copy, because this knowledge has provided a framework for achieving the best copy possible. We'll explore some of the ramifications of this breakthrough in the next section. Next section: Applying the Spectrum |
