API Introduction

ISP tunning algorithm APIs are provided for developers to perform ISP secondary development and image quality tunning. Only the Atlas 200I/500 A2 inference products supports functions in this part.

The ISP supports standard image data processing of sensors, including basic functions such as automatic white balance (AWB), automatic exposure (AE), demosaic, defect pixel correction (DPC), and lens shading correction (LSC) as well as advanced functions such as sensor wide dynamic range (WDR), dynamic range compression (DRC), and denoising. The following table describes the image processing functions supported by the ISP.

**Table 1** ISP algorithm functions
Algorithm	Function	Description
BLC	Black level correction	Provides sensor-related BLC.
DPC	Defect pixel correction	Detects and corrects static and dynamic defect pixels.
BNR	Bayer noise reduction	Implements image NR in the Bayer domain to retain details while reducing image noise.
FPN	Fixed pattern noise (FPN) removal	Corrects the image input from the sensor based on the calibrated black frame or black line to remove the FPN of the sensor.
DEMOSAIC	Demosaicing	Converts raw images in Bayer format into RGB images.
CAC	Purple fringing correction	Corrects the axial chromatic aberration (purple fringing) and lateral chromatic aberration (color fringing on opposite sides of an object with different colors) introduced by the lens.
GAMMA	Video luminance adjustment	Adjusts the luminance through the R, G, and B channels based on the gamma curve.
DRC	Dynamic range compression	Adjusts the display dynamic range of the image so that the display effect on the display device is consistent with the perception of human eyes.
WDR	Wide dynamic range	Provides the frame merging WDR functions: Sensor built-in WDR 2-to-1 WDR at most
AWB	Automatic white balance	The spectral components of visible lights vary according to the color temperature. The white objects have a red cast at a low color temperature or have a blue cast at a high color temperature. Human eyes can identify the actual object color based on brain reflections. The automatic white balance (AWB) algorithm is used to reduce the impacts on the actual object color exerted by external illuminants. This ensures that the captured color information is converted into the information without color cast under an ideal illuminant.
AE	Automatic exposure	Collects AE statistics and implements AE through sensor adjustment based on statistics.
AF	Auto focus	Collects AF statistics and image clarity evaluation information, and implements AF based on the information. Currently, only the AF statistics can be collected, whereas the AF algorithm is not supported.
LSC	Lens shading correction	Corrects lens shading. Currently, only mesh shading is supported.
SHARPEN	Image sharpening	Implements image sharpening to improve the image definition.
DEHAZE	Automatic dehazing	Provides the powerful dehazing function based on regions to improve video contrast and definition in the haze.
CLUT	3D color lookup table (CLUT) enhancement	Implements complex color adjustment with the 17 × 17 × 17 3D LUT, for example, adjustment of luminance, saturation, shade regions, medium-luminance regions, and highlighted regions.
LDCI	Local contrast enhancement	Enhances local contrast based on local histogram equalization to improve detail effect in dark regions, and enhances the high frequency in images to improve contrast.
CA	Color adaptation	Adjusts the saturation and implements thermal imaging coloring. Currently, thermal imaging coloring is not supported.
3DNR	3D noise reduction	Removes Gaussian noise from images through parameter configuration, to smooth the images and reduce the encoding bit rate.
HLC	High luminance shielding	Grays the region with the highest luminance in an image to reduce the stimulation of unpleasant luminance such as car headlights to human eyes.
GE	Image quality improvement	Equalizes the Gb and Gr channels when imbalance occurs and improves the image quality in some scenarios.
Expander	Data decompression	Decompresses the compressed data in the sensor.
DG	Digital gain	Provides digital gains for each channel.
MG	Maximum gain statistics	Collects statistics on the average value of divided blocks after the DRC, and compares the average value with that of the AE statistics to obtain the maximum gain. The MG statistics include the average value statistics on the blocks R/Gr/Gb/B based on 8-bit precision. The maximum number of blocks is 17 × 15.
CCM	Color correction	Linearly corrects the color space by using the standard 3 × 3 matrix and vector offset.
CSC	Color space conversion	Coverts RGB into YUV by using the standard 3 × 3 matrix and vector offset.
DIS	Digital image stabilization	Calculates the inter-frame offset by comparing the difference between two frames and then crops the image accordingly to eliminate the image jitter.
LDC	Lens distortion correction	Performs lens distortion correction and spreading on an image frame. Currently, the spreading function is not supported.
FISHEYE	Fisheye correction	Performs fisheye correction on an image frame (including 360° panoramic, 180° panoramic, and normal correction modes).