On isp-ccm

One ． summary

    CCM yes ISP pipeline  Important module about color adjustment in , Introduce CCM First understand the color sensitivity of human eyes and semiconductors .

There are three types of cone cells in the retina of the human eye (cone) For feeling blue 、 green 、 Red color light of three frequencies , A kind of rod cell (rod) Only sense brightness information under low illumination conditions . The absorption rate of green light by human eyes can reach 20%, The absorption rate of orange light is 4% about , The highest absorption rate of blue light is only 2% about . The human eye is right 480~640nm The hue resolution of interval color light is high , among , Yes 500nm ( Turquoise ) and 600nm( Orange yellow ) For two wavelengths , As long as the wavelength changes by about 1nm, You can distinguish the change of tone . It can be seen that there are cfa There are two G Or two Y The filter of is reasonable （RGGB,RYYB）.

     For semiconductors , A typical silicon material sensor stay 500nm Place to blue 、 The response of green light is almost equal , But the blue cones of the human eye are right 500nm The blue-green light response of is almost zero , There is a big difference between the two . At the same time, it is made by different manufacturers sensor It will be different , For different bands R,G,B It will be different .

     besides , Other optics , For example, lens , lens coating ,IRCUT The frequency response of will cause deviation to the color restoration . Therefore, it is necessary to correct the color obtained by the synthesis of all optical devices , To match the perception of the human eye .

Two ．CCM correct

    CCM Is in RGB What is done in the domain , The mainstream way is to use a 3*3 The matrix of will be input RGB Map to a new pixel value R’ G’B’.

In order to maintain awb White balance of calculation results , Need make a00+a11+a22=1. There are tens of millions of combinations of input colors , and CCM The parameters are only 9 individual , In fact, only a few of the most important colors can be prioritized to meet the requirements of the human eye , It is impossible to guarantee that all colors are optimal under all conditions . Later, other algorithm modules can be used to replace or supplement, such as ：CLUT, The way of color mapping , Realize the adjustment of the mapping of thousands of colors .

     In practice, we usually use X-Rite 24 On the color card 18 Calculate the correction coefficient with color patches as the standard , That is, use the color card under different color temperatures , After getting the image , extract 18 The average color of each color block , Form the input matrix S, Use standard RGB Value as target S’：

S'=M·S

Usually solved directly M The least square method can be used , But there are a00+a11+a22=1 when , We need to use the least square method with constraints to solve .

Of course, sometimes in order to meet the needs of some colors ： Like a beautiful face 、 Personality tone 、 Customer preferences , It needs to be adjusted manually CCM matrix . But at this time , What we see RGB The image is through AWB、CCM、Gamma After nonlinear treatment , As shown in the following formula . So when inching （ Big move ） The matrix may have larger （ Fretting ） Fluctuation , meanwhile ,CCM Matrix needs and Gamma binding , When gamma After change ,CCM It needs to be readjusted .

For more information, please pay attention to my official account ： I love image processing