What is white balance? How to understand the white balance?

First of all to understand what is the white balance, it refers to the process of image processing, the original material is a white object of the image color reduction, the effect of removing the external light color temperature, so that it is also shown in the photo white.

How does the color temperature understand? This concept is actually a bit of a detour, it is the Kelvin (blackbody) This ideal light source, at different temperatures, the emitted light color characteristics to define. The blackbody is an idealized concept. It is an object that glows, but absorbs any light that comes from the outside, and at the same time it completely releases all the energy absorbed into the form of light, so it is called a blackbody. The temperature sheet of the blackbody is called the Kelvin (K). In the case of temperature changes from 3300K to 9300K, its luminous colors are red, white, and blue, respectively. Easy to understand, you can imagine, the flame you must have seen, cone temperature is low, the Wei temperature is higher (blue), the cone temperature is yellow (warm), so the most outside of the flame temperature is blue (cool), is not a bit anti-common sense. and the color temperature is the use of this temperature variation of the blackbody to quantify the color trend. Color temperature low value, partial yellow, color temperature high value, blue, so-called warm and cold tones is a more perceptual term.

<3300k

Warm

3300~6000k

White

>6000k

Cool

The following are the color temperatures for several common scenarios:

Candle light 1930K, tungsten filament 2900K, noon Sun 5600K, Blue Sky 18000K

Before talking about how white balance is fixed, there are two important theories to know:

Gray World Theory: This theory is interesting, rather than color science, I think more like the category of statistics, it is believed that any image, when there is enough color change, its R, G, b component mean value will tend to balance (that is, RGB three values equal, that is, black and white gray type of color). This theory is widely used in the global white balance, which is characterized by the ability to use more image information to make judgments, but in the face of a single color of the image is somewhat weak.

Total reflection Theory: The most luminance point in an image is the white dot, which is assumed to be the largest Y-value point in the YCbCr space, which corrects the entire image. Features are only considered the brightest part of the color, with the above the Gray world theory is the opposite, in the processing of color partial monotonous image effect is better, but the face of color-rich pictures, because the brightest point is not necessarily white, may appear biased color situation.

The two theories correspond to the theoretical basis of two color space RGB and YCbCr adjusting white balance: Judging whether a picture white balance is accurate, if inaccurate, how to quantify its deviation value.

White balance is a process to correct the overall color of the picture, then why does the human eye not need it? In fact, when you see a thing, the eye has already made color correction to it. The camera certainly has no human eye so smart (at least not now), in summary, its automatic white balance algorithm is to set a range, if the color of the picture taken by the average fall in this range, then OK, it does not need to fix. If you deviate from this range, you need to adjust the parameters and correct the color data until the average value falls within the specified range. This is the WB white balance correction process.

The following is a brief introduction of a few white balance algorithm approximate principle, but the specific gain calculation and mathematical modeling process is not to repeat, this thing normal people see the head of the big.

1. Gray World Algorithm (assumption)

The principle of this algorithm is very simple, according to the previous theory of gray-scale world, the original image of the RGB is adjusted to r=g=b. Imperfect place is this algorithm on the color of the image is not rich in general sensitivity, the processing effect will not be ideal, the limitations are larger.

2. Standard deviation weighted gray world algorithm (deviationweighted Gray worlds assumption)

Standard deviation weighted gray world algorithm is aimed at the improvement of the previous algorithm, its principle is to divide the image into a few pieces, and then the use of statistical analysis of each block, look at the richness of color, color is weighted, the color is less weight, the final sum to obtain a mean value. The RGB values are corrected based on this relatively accurate value.

3. Full Reflection Algorithm (Perfect Reflector assumpution)

This is based on the previous introduction of the theory of total reflection generation algorithm, it is not difficult to understand, it is believed that the brightest point is white, if the original image of the brightest point is not, then the deviation from the white value of the inverse correction. The disadvantage is that if the image color is complex or there is no high-light point, its correction effect will be relatively weak.

The above three algorithms are relatively simple, the computation amount is not big, but each has the advantages and disadvantages, also further derived more effective but more complex mixed algorithm, for example: Brightness weighted gray world algorithm and total reflection algorithm orthogonal combination algorithm (quadraticcombining luminance Weighted Gray World & Prefect Reflector Assumption). Look at this long name to know, this algorithm is very complex, white balance correction effect is pretty good, and it is convergence, in the image processing time will not bring too much loss, but unfortunately the computational volume is huge, the requirements of hardware resources too high.

Algorithm is very boring, non-science and technology is not interested in mathematics can also be directly skipped, generally know this is the right thing to do. But I can give the conclusion is: the better the final effect of the algorithm, its complexity is higher, the greater the computational capacity, the requirements of the hardware circuit is high. The concrete implementation also needs to weigh in the white balance correction ability, the algorithm execution efficiency, the processor hardware performance three aspects.

And through the above understanding you will find that if the ISP image processor can be high, the white balance algorithm will be much more space to cast. Some times the white balance is not accurate, to a certain extent, and ISP performance is not related, of course, here also depends on the manufacturers in the algorithm optimization of the foundation. In general, the same generation of various brands of the flagship processing performance differences will not be particularly large, although the software is not easy to see the effort, but the importance is undoubtedly, whether the overall performance of the hardware fully play out is the key.

For example, in the era of DSLR just digitized, the camera's image processor performance is relatively low, difficult to withstand the high computational strength of the white balance algorithm, so many SLR cameras (such as Canon 1 D, Nikon D2, Olympus E-1, etc.) on the fuselage has a white balance induction device (that is, the small white dot on the front of the fuselage) ), this can help improve the white balance accuracy. As the performance of the camera image processor soared, presumably from Fujitsu's Expeed generation processor to the Nikon OEM, the device was removed from the external white balance sensor. More accurate color temperature correction with increasingly powerful processors with more and more RGB metering partitions. More than one sentence here, the more partitions, the more accurate white balance sampling processing, but also will bring the amount of computational overhead, from the initial only a few partitions, to the D800 above the 91,000-megapixel RGB sensor, while the metering and white balance calculation at the same time, even the strength of the face recognition, the back The Expeed 3 is the biggest contributor to the ARM architecture. And like DC ah, mobile phone camera ah this kind of continuous camera, is used in the previous frame image processing results applied to the back of the image, the implementation of the SLR above the individual metering sensor is not the same. This is caused by the difference in the structure of the product itself.

In this set of images, it is taken under different white balances:

Source: www.zealer.com Li Yu HTTP://WWW.ZEALER.COM/QUESTION/4

What is white balance? How to understand the white balance?