One of the ISP image processing algorithms black level correction

The past year due to work, has been busy, there is no time to update some ISP algorithm, after a year of efforts, their own ISP algorithm also has a deeper understanding, so decided to recently to the ISP algorithm to share some of the understanding, for everyone to reference the study, hope to engage in this line of people useful.

This paper mainly introduces one of the algorithms of ISP pipeline black level correction:

It is well known that the information collected by CMOS sensors is transformed by a series of transformations, and the raw data is generated in raw raw format, and only one color information per pixel is available, but only one RGB, but because the frequency of the sensor to the green light can all pass, so raw data inside, There are two G one r, and one B. To get the RGB data, we have to use the information we have to obtain the missing information, which is described later. Here goes the black level correction said.

We know that image data is generally 0--255, but sensor in the factory, manufacturers will generally set the image data output range such as 5-250, anyway, the lowest level is not zero. Therefore, we need to adjust the image data range so that its minimum value is zero, which is the black level correction.

Black level correction There are two methods: one is directly on the raw data to subtract a value, can be RGB minus the same value, can also be different, most manufacturers such as, HiSilicon, such as the basic use of this practice. One is the use of a function, this is more troublesome, it is not introduced.

PS: This is the simplest approach in the ISP and will introduce some other complex algorithms in detail later.

One of the ISP image processing algorithms black level correction