I was thinking about how the white balance algorithm would be implemented before the equation was shown while looking at the examples, and the actual equation is scarily simple.
I was thinking just fixing the white balance would wildly change the other colors negatively, but it maps to very natural colors.
Eloyye
I think the lecture brought up an interesting point about the problem of calculating the raw input of white object in order to create an automatic white balance output. For RGB values I believe the range of values are from [0, 255] but do we make that assumption here?
LeslieTrue
Other than RGB, there are also some other representations for colors. ChatGPT provides me with these information which can be an interseting reference.
CMYK: This color model is used in printing and stands for cyan, magenta, yellow, and key (black). The combination of these colors produces a wide range of colors, with black being used to enhance the contrast and depth of the image.
HSL and HSV: HSL (hue, saturation, and lightness) and HSV (hue, saturation, and value) are two color models used in computer graphics and image processing. They allow for easier manipulation of the hue and saturation of a color, as well as the brightness or lightness.
Lab: This color model is used in scientific applications and represents colors in terms of their lightness, a and b components. The lightness component represents brightness, while the a and b components represent the color in terms of green-red and blue-yellow axes.
YUV: This color model is used in video encoding and transmission and separates the color information into luminance (Y) and two chrominance (U and V) components. The luminance component represents the brightness of the color, while the chrominance components represent the color in terms of the blue and red color differences.
kkkhanl
Did we have to sample white objects from the image to get R'w, G'w and B'w?
starptr
Is this the same strategy modern screens use to adjust to the environmental light colors? Some of the newer smartphones' screens adjust colors and sometimes I don't notice it until the camera of the phone is in a different light than the rest of the screen.
Staffyirenng
@kkkhanl The algorithm needs to infer or guess at what RW′, GW′ and BW′ would be in the scene, even if white is not present. This is a challenging, ill-posed problem for many photographs.
Simple AWB algorithms make overly simplified assumptions, such as: that the average color of a photo is 18% gray, or that the brightest object in the scene is white. More sophisticated algorithms may look for certain objects in the scene, such as faces, and priors on those scenes, such as a database of skin-tones, in order to help infer the likely illuminant color and therefore the color of white.
This is a cool paper with a sophisticated white balance algorithm.
I was thinking about how the white balance algorithm would be implemented before the equation was shown while looking at the examples, and the actual equation is scarily simple.
I was thinking just fixing the white balance would wildly change the other colors negatively, but it maps to very natural colors.
I think the lecture brought up an interesting point about the problem of calculating the raw input of white object in order to create an automatic white balance output. For RGB values I believe the range of values are from [0, 255] but do we make that assumption here?
Other than RGB, there are also some other representations for colors. ChatGPT provides me with these information which can be an interseting reference.
CMYK: This color model is used in printing and stands for cyan, magenta, yellow, and key (black). The combination of these colors produces a wide range of colors, with black being used to enhance the contrast and depth of the image.
HSL and HSV: HSL (hue, saturation, and lightness) and HSV (hue, saturation, and value) are two color models used in computer graphics and image processing. They allow for easier manipulation of the hue and saturation of a color, as well as the brightness or lightness.
Lab: This color model is used in scientific applications and represents colors in terms of their lightness, a and b components. The lightness component represents brightness, while the a and b components represent the color in terms of green-red and blue-yellow axes.
YUV: This color model is used in video encoding and transmission and separates the color information into luminance (Y) and two chrominance (U and V) components. The luminance component represents the brightness of the color, while the chrominance components represent the color in terms of the blue and red color differences.
Did we have to sample white objects from the image to get R'w, G'w and B'w?
Is this the same strategy modern screens use to adjust to the environmental light colors? Some of the newer smartphones' screens adjust colors and sometimes I don't notice it until the camera of the phone is in a different light than the rest of the screen.
@kkkhanl The algorithm needs to infer or guess at what RW′, GW′ and BW′ would be in the scene, even if white is not present. This is a challenging, ill-posed problem for many photographs.
Simple AWB algorithms make overly simplified assumptions, such as: that the average color of a photo is 18% gray, or that the brightest object in the scene is white. More sophisticated algorithms may look for certain objects in the scene, such as faces, and priors on those scenes, such as a database of skin-tones, in order to help infer the likely illuminant color and therefore the color of white.
This is a cool paper with a sophisticated white balance algorithm.