Note

This is the documentation for the latest development branch and may refer to features that are not available in released versions. If you are looking for the documentation for a specific release, use the drop-down menu on the left and select the desired version.

K230 ISP Image Tuning Guide

Note: This document focuses on the complete ISP tuning methodology and module parameter descriptions. For the hands-on BLC/LSC/CC calibration workflow, also refer to: how_to_calibrate_isp.md.

Overview

When tuning image quality, the primary areas of focus are: brightness, color, contrast, and sharpness. Through combined sensor/lens calibration and ISP sub-module tuning, the goals are:

• Overall image brightness at a reasonable level

• Accurate color reproduction

• Good sharpness with no obvious noise

• High contrast — an overall transparent, vivid look

Overall image tuning flow:

K230 ISP pipeline:

Important notes:

• In K230 HDR mode, the CSI module does not support sensor data patterns where Hsync precedes Vsync. The Vsync pulses for each frame (L/S/VS) must not overlap during the active data period.

• When TNR (in 3DNR) is enabled, the ISP requires that sensor Hblank ≥ 180 pixel clocks (256 recommended), and Frame length – Active lines ≥ 92 lines.

---

Calibration

Overview

The calibration tool (K230ISPCalibrationTool.exe) calibrates parameters for six ISP modules: BLC, LSC, CC, AWB, Noise Profile, and CAC.

Calibration order:

Prerequisites: install MATLAB Runtime (R2023a) before using the calibration tool. Download: https://ssd.mathworks.com/supportfiles/downloads/R2023a/Release/0/deployment_files/installer/complete/win64/MATLAB_Runtime_R2023a_win64.zip

Main interface:

Click the module button to jump directly to that module’s calibration interface.

Black Level Correction

Principle

When an analog signal is very weak, the A/D converter may fail to capture it, causing loss of detail in dark areas. To address this, the sensor adds a fixed offset to the analog signal before A/D conversion, preserving more image detail in the output digital signal. Black level correction determines this offset value precisely. All subsequent ISP modules must subtract this offset first to maintain data linearity.

Capture RAW Images

1. Cover the lens completely in a fully dark environment — no light must reach the sensor.

2. Set the sensor to manual exposure mode.

3. Capture 12 RAW images across different gain (1×, 2×, 3×, 4×) and integration time (0.01 ms, 0.02 ms, 0.03 ms) combinations.

4. Save to folders named by the combination, e.g. Gain_4_T_0.03 (gain=4, time=0.03 s).

Run Calibration

Click Black level correction in the main interface:

Steps:

1. Click Select and choose the RAW image folder.

2. Align Pos field: enter the Bayer pattern of the RAW images.

3. Set input/output bit width.

4. Set resolution (width × height).

5. Click OK to start calibration.

Options:

• Combine Channel: if checked, BLC is the combined measurement for all four R/Gr/Gb/B channels.

• Normal File Out: outputs blc_para.txt.

• Extended File Out: outputs per-channel BLC at each exposure/gain combination separately.

Example calibration result chart:

Lens Shading Correction (LSC)

Principle

LSC eliminates the darkening towards the image corners caused by uneven optical refraction through the lens. The brightness fall-off follows a cosine-to-the-fourth-power law. Calibration captures this attenuation trend and stores the correction in a 32×16 grid to restore each region’s brightness to the center reference level. Both luma shading and color shading are calibrated. Because color shading characteristics differ at different color temperatures, LSC must be calibrated separately for each target color temperature.

Capture RAW Images

Precautions:

• Adjust AE target brightness so the image center average is about 80% of the maximum (e.g. 204/255 for 8-bit).

• Use a flat, uniform light source with a smooth, texture-free target — a DNP light box is ideal.

Steps:

1. Aim the lens at the DNP light box target area; ensure no environmental interference.

2. Turn on the light box, select D65, set appropriate illuminance.

3. Capture one RAW image.

4. Switch to the next light source (D50, TL84, F12, A) and repeat.

Run Calibration

Click Lens Shade correction in the main interface:

Steps:

1. In region 2: select the Bayer pattern, enter width/height/bit-width, choose the correction type (Color Shading or Luma+Color Shading), select the color temperature in the CCT dropdown, and enter the Black Level Offset for each channel.

2. Click Load image in region 1 to import the RAW image to calibrate.

3. In region 3: set the grid knots. K230 ISP LSC hardware uses a 32×16 grid → choose “33Full × 17Symm knots”. If using ALSC, check “ALSC uniform…”. Note: if “Automatic initial knot positioning” is checked, ALSC cannot be enabled in auto.json.

4. In region 4: optionally set the max correction ratio for center/corners.

5. In region 5: set the corner compensation ratio (e.g. 80% means corners are corrected to 80% of center brightness).

6. Leave other settings at defaults.

7. Click Start to calibrate; save the LSC data.

8. In region 6: apply the LSC data to the image to preview the result; click Save image to file to save a PNG for each light source.

When setting the compensation ratio in step 5, consider how severe the lens shading is. Heavy shading requires large corner gains, which can amplify corner noise. Reduce the compensation ratio to balance noise in the corners.

Color Correction (CC)

Principle

CCM calibration measures the actual colors of 24 color checker patches as captured by the sensor and compares them to expected (reference) values to compute a 3×3 CCM matrix. This matrix transforms the sensor’s color space into the standard sRGB color space.

Capture RAW Images

Steps:

1. Place the 24-patch color checker inside or in front of the standard light box (lens facing the checker or at 45°).

2. Turn on the light box, select D65, and adjust illuminance.

3. Adjust distance and position so the color checker occupies approximately 2/3 of the image.

4. Adjust AE target so the RAW average brightness is around 50.

5. Capture one RAW image.

6. Switch to each other light source (D50, TL84, F12, A) and repeat.

7. With the lens position unchanged, remove the color checker and capture background RAW images for all five light sources.

Run Calibration

Click Color Correction in the main interface:

Steps:

1. In region 1: set width, height, bit-width, Black Level Offset, and Bayer pattern.

2. Click Load to import:

   • Load sRGB References — CC_Standard.cxf (in Canaan.Calibration.Samples\Data\)

   • Load Color Checker Image — color checker RAW

   • Load Background Image — background RAW

   • To use custom Lab references: Load Lab-Customer References; or load a reference image and use Function → Select Reference Color Check.

3. Click LSC and load the calibrated LSC parameter file.

4. Configure calibration parameters: set gamma, check CIELAB reference, configure 24-patch weights, set preferred light source, set output saturation (default 1).

5. In region 3: select the color patch detection mode (automatic, semi-automatic, or manual). In semi-automatic mode, use the crosshair to click the four corner patches in order.

6. Click Calibrate to start.

Auto White Balance (AWB)

Principle

AWB calibration computes the optimal Planckian locus fit and color temperature fit curve based on the sensor’s white point characteristics (R/G, B/G) under several standard light sources. The goal is to enable the camera to automatically adapt to different color temperature conditions for accurate white and color reproduction.

Run Calibration

Click Auto White Balance in the main interface:

Steps:

1. Click file and load the sensor spectral sensitivity file:

   

   If you don’t have your sensor’s spectral sensitivity file, use the default OV2775_sensitivity.txt (in Canaan.Calibration.Samples\Data\). AWB calibration no longer strictly depends on this file, so any valid file works.

2. Click illumination and load CIE_Illuminants.cxf (in Canaan.Calibration.Samples\Data\). Select at least 3 light sources, click OK, and categorize them as indoor or outdoor using the dropdown.

   

   

3. Click AWB V2+ Calibration (or Start Calibration in older versions) and select the CC calibration output files for each light source:

   

4. Check White Points Evaluate and select an image for white point estimation.

5. Click Gain Region Modifier, select the PNG images saved during LSC calibration for each light source, and click OK:

   

6. Define the initial size of the gain polygon; click OK:

   

7. Manually adjust the orange polygon (near-white zone) and black polygon (all pixels inside are classified as white) until the black polygon covers all selected light source white points; save the data:

   

K_Factor Calibration

In the AWB algorithm, the outdoor condition is: Exp × K_Factor ≤ 0.12 (where Exp is the exposure value).

For example, if 2000 lux is the outdoor/transition illuminance threshold: obtain the exposure value (ET × gain) at that illuminance, then: K_Factor = 0.12 / (ET × gain).

A larger K_Factor indicates higher sensor sensitivity; a smaller value indicates lower sensitivity. This parameter is in the AWB section of the XML file.

Noise Calibration

Principle

Noise profile calibration quantitatively measures and characterizes the noise produced by the sensor under different conditions (light levels, temperature, ISO). Understanding these noise characteristics is key to effective noise reduction.

Capture RAW Images

1. Place a black-white gradient test chart inside the standard light box; select D50.

2. Disable AE and switch to manual exposure. Capture two sets of 30 RAW images each:

   • Bright set: adjust light box brightness so the dark area of the gradient chart is at a medium gray value.

   • Dark set: adjust light box brightness so the white area of the gradient chart is at a medium gray value.

   • Both sets: keep the same exposure time and gain at 1×.

Run Calibration

Click Noise Calibration in the main interface:

Steps:

1. Set RAW width, height, bit-width, and Black Level Offset.

2. Click file to import both the bright and dark RAW image series.

3. Click ROI to define a region of interest if needed.

4. Adjust parameters: for the low-light histogram, use the range from Lower Range to Split Point; for the highlight histogram, use Split Point to Higher Range. The two combined should cover nearly all 0–255 values (for 8-bit). The preserved regions of Highlight and Lowlight should follow a Poisson distribution. If BLC offset is set, Lower Range ≥ BLC value.

5. Leave Gamma Correction and Model Fitting at defaults.

6. Click Evaluate Measures, then click Fit to Model to complete the calibration and save.

Chromatic Aberration Correction (CAC)

Principle

Chromatic aberration is a common optical defect where the lens cannot focus all wavelengths at the same point, causing red/green or blue/purple fringing at image edges. CAC calibration builds a mathematical model from captured image data to describe the aberration, then determines correction parameters.

Capture RAW Images

1. Place a checkerboard or dot pattern chart inside the standard light box; select D50.

2. Adjust AE so the minimum value in black patches is greater than the BLC value, and the maximum value in white areas does not exceed 208.

3. Capture one RAW image and save it.

Run Calibration

Click Chromatic Aberration Correction in the main interface:

Steps:

1. Click Demosaic&Filter and set the Bayer pattern.

2. Click Load image to load the RAW image.

3. Configure the Demosaic&Filter settings:

   • ISP_demosaic_threshold: default 4. Higher values detect fewer edges; lower values detect more. 255 disables edge detection.

   • Simulate 2 additional Line Buffers: expands vertical offset range to ±3; no chroma down-filter is applied. Off by default.

   • Filter stage1 select, Sharpen Level, Denoise Level: set as needed.

   • Lum_weight Off: disables the noise weighting function (helps in dark areas). Off by default.

4. Click Color Correction and configure:

   • Check ISP_bls_enable and set the correct BLC.

   • Check Auto White-Balance.

   • Set CCM.

   • Set global gain to 1.0.

5. Click Gamma and set the gamma curve (default: gamma 2.2).

6. Click LSC → Load parameters, load the D50 LSC calibration data, and check Enable.

7. Click CA Estimate, check CA_Estimate, and click Process to run.

8. Click CAC to view the calibration results.

---

Module Reference

BLS

Description: Subtracts the black level offset to maintain data linearity.

Parameter	Type / Range	Description
bls_enable	bool	BLS enable switch
bls	int[4], 0–4095	Black level offset for four channels (based on ISP 12-bit)

Tuning note: Although four independent channel values are supported, using the same value for all four is recommended.

---

LSC

Description: Corrects lens shading using a 32×16 grid applied to the four RAW channels.

Parameter	Type / Range	Description
enable	bool	LSC enable switch
matrix	float[4][1089]	Lens shading correction matrices for R, Gr, Gb, B; from calibration file
x_size	int[32]	Distance between adjacent X-axis grid knots; from calibration file
y_size	int[16]	Distance between adjacent Y-axis grid knots; from calibration file

---

Dgain

Description: ISP digital gain, primarily used to increase image brightness.

Parameter	Type / Range	Description
enable	bool	Digital gain enable switch
digital_gain_r	float, 1.0–255.99	R channel gain
digital_gain_gr	float, 1.0–255.99	Gr channel gain
digital_gain_gb	float, 1.0–255.99	Gb channel gain
digital_gain_b	float, 1.0–255.99	B channel gain

---

AE

Description: Auto Exposure controls image brightness. Key tuning areas: AE target brightness, convergence speed, and smoothness.

Parameter	Type / Range	Description
enable	bool	AE enable switch
antiFlickerMode	int, 0–3	Anti-flicker mode: 0=Off, 1=50 Hz, 2=60 Hz, 3=User defined
autoHdrEnable	bool	In HDR mode: true=auto-compute HDR ratio; false=use fixed HDR ratio
dampOver	float, 0–1.0	Damping factor for smooth convergence when overexposed
dampOverGain	float, 0–128.0	Convergence acceleration gain when overexposed and out of clip range
dampOverRatio	float, 1.0–4.0	Clip range ratio for overexposure; smaller = faster convergence
dampUnder	float, 0–1.0	Damping factor for smooth convergence when underexposed
dampUnderGain	float, 0–16.0	Convergence acceleration gain when underexposed and out of clip range
dampUnderRatio	float, 0–1.0	Clip range ratio for underexposure; larger = faster convergence
expV2WindowWeight	float[32×32], 0–255	Per-zone exposure weight
frameCalEnable	bool	Enable frame interval for exposure setting
lowLightHdrGain	float[20], 0–255.0	Gain values per gain step in WDR mode
lowLightHdrLevel	int, 0–16	Total gain steps in WDR mode
lowLightHdrRepress	float[20], 0–1.0	Target brightness suppression ratio per gain step in WDR mode
lowLightLinearGain	float[20], 0–255.0	Gain values per gain step in linear mode
lowLightLinearLevel	int, 0–19	Total gain steps in linear mode
lowLightLinearRepress	float[20], 0–1.0	Target brightness suppression ratio per gain step in linear mode
motionFilter	float, 0–1.0	Smoothing parameter for motion factor in adaptive scene evaluation
motionThreshold	float, 0-1.0	Motion detection threshold
roiNumber	int	Current ROI window index
roiWeight	float	Brightness calculation weight for the current ROI window
roiWindow	float (fx,fy,fw,fh)	Current ROI start coordinates (x,y) and size (w,h)
semMode	int, 0-2	Scene evaluation mode: 0=off, 1=fixed, 2=dynamic
setPoint	float, 0–255.0	AE target brightness
targetFilter	float, 0–1.0	Smoothing factor for AE target changes; larger = faster changes
tolerance	float, 0–100.0	AE target brightness lock range (%)
wdrContrast.max	float, 0–255.0	Max contrast for AE setpoint in adaptive scene evaluation
wdrContrast.min	float, 0–255.0	Min contrast for AE setpoint in adaptive scene evaluation

---

AWB

Description: Reduces the influence of ambient light on object color, converting color-shifted objects to appear as they would under ideal daylight.

Parameter	Type / Range	Description
enable	bool	AWB enable switch
awbTempWeight	float, 0.0–1.0	Color temperature light source weight
mode	int, 0 or 1	0=AWB, 1=AWB METADATA
roiNumber	int	Current ROI window index
roiWeight	float	AWB calculation weight for the current ROI window
roiWindow	float (fx,fy,fw,fh)	Current ROI start coordinates (x,y) and size (w,h)
useCcMatrix	bool	Enable adaptive CCM
useCcOffset	bool	Enable adaptive CCM offset
useDamping	bool	Enable AWB damping
kFactor	float	Photo-sensitivity coefficient for outdoor/transition classification

kFactor calibration: outdoor condition: Exp × kFactor ≤ 0.12. Example: at 2000 lux as the outdoor threshold, kFactor = 0.12 / (ET × gain). Larger kFactor → higher sensor sensitivity; smaller kFactor → lower sensor sensitivity. This parameter is in the AWB section of the XML file and must be calculated using the method above before filling in.

---

WDR

Description: Wide Dynamic Range — improves image contrast using histogram equalization, recovering detail in both highlights and shadows.

Parameter	Type / Range	Description
enable	bool	WDR enable
contrast	int, -1023–1023	Higher = stronger local contrast
entropy	int[20]	Local weight
entropy_base	int	Brightness factor: larger base → stronger local contrast
entropy_slope	int	Brightness factor: larger slope → weaker local contrast
flat_strength	int, 0–19	Flat region stretch strength
flat_thr	int, 0–20	Flat region threshold; below = flat, above = textured
gamma_down	int[20]	Local weight
gamma_pre	int[20]	Local weight
gamma_up	int[20]	Global curve
global_strength	int, 0–128	Global contrast strength
high_strength	int, 0–128	Highlight protection; larger = stronger highlight preservation
low_strength	int, 0–255	Shadow protection; larger = stronger shadow preservation
strength	int, 0–128	Overall strength

Tuning note: In linear mode, set strength=128. low_strength is effectively the max gain — increase it to brighten shadows. high_strength suppresses overexposed highlights. As gain increases, gradually reduce low_strength to avoid noise amplification. Adjust contrast, flat_strength, and flat_thr for local contrast.

---

GE (Green Equalization)

Description: Balances Gr and Gb pixel differences in the RAW data to prevent checkerboard or maze-like artifacts from demosaic interpolation. This module is placed before DPCC.

Parameter	Type / Range	Description
enable	bool	Green equalization enable
threshold	float, 0–511.0	Green equalization strength threshold

---

DPCC (Defective Pixel Cluster Correction)

Description: Corrects defective pixels before the demosaic module. The static bad-pixel table supports up to 2048 entries. Seven correction setting groups are available.

Parameter	Type / Range	Description
enable	bool	DPCC enable (default off)
bpt_Enable	bool	Bad pixel table enable
bpt_Num	int, 0–1024	Index of current bad pixel in table
bpt_out_mode	int, 0–14	Median output mode for bad pixel correction
bpt_pos_X	int	Horizontal address of bad pixel
bpt_pos_Y	int	Vertical address of bad pixel
bypass	bool	Bypass DPCC
line_mad_fac	int[2][3], 0–63	Mean absolute difference factor
line_thresh	int[2][3], 0–255	Row threshold for R/G/B
methods_set	int[3], 0–8191	Bad pixel correction method selection per channel
out_mode	int, 0–15	Interpolation mode for correction unit
pg_fac	int[2][3], 0–63	Peak gradient factor
rg_fac	int[2][3], 0–63	Rank gradient factor
rnd_offs	int[2][3], 0–3	Rank neighbor difference rank offset
rnd_thresh	int[2][3], 0–63	Rank neighbor difference threshold
ro_limits	int[2][3], 0–3	Rank offset limit
set_use	int, 1–7	Select which correction setting group to use

---

DPF (Bilateral Filter)

Description: Bilateral filter for noise reduction.

Parameter	Type / Range	Description
enable	bool	Enable switch
gain	float, 1.0–1000.0	Sensor gain
gradient	float, 0.1–128	Gradient
min	float, 1.0–128.0	Strength clip
offset	float, 0–128.0	Higher = stronger noise reduction
div	float, 0–64.0	Higher = weaker noise reduction
sigma_g	float, 1.0–128.0	Spatial filter for G channel; higher = more weight
sigma_rb	float, 1.0–128.0	Spatial filter for R/B channels; higher = more weight
noise_curve	float, 0–4095.0	17-point curve; from calibration file

---

DNR (3D Noise Reduction)

Description: 3DNR combines temporal and spatial domain information for more effective noise reduction. Motion detection ensures moving objects retain sharpness.

Parameter	Type / Range	Description
enable	bool	Master enable for 2DNR and 3DNR
tnr_en	bool	3DNR enable
nlm_en	bool	2DNR enable
sigma	float, 0.1–16.0	2DNR strength; higher = stronger
strength	float, 0–128.0	2DNR strength; higher = stronger
blend_motion	float, 0–100.0	Blend weight between 2DNR-processed and TNR-output frames; affects denoise strength without changing motion estimation
blend_slope	float, 0.1–32.0	Blend slope; higher = more weight to NLM. Smaller = higher static blend weight; larger = higher motion blend weight
blend_static	float, 0–100.0	Blend weight for static regions between 2DNR and TNR output
dialte_h	float	Motion dilation window width
filter_len	float, 0–1024	Temporal filter window length (number of reference frames)
filter_len2	float, 0–100	Motion history frame count
motion_dilate_en	bool	Motion dilation enable
motion_erode_en	bool	Motion erosion enable
noise_level	float, 0–1024	Motion detection threshold; above = motion region
noisemodel_a	float	Slope of custom pre-gamma curve
noisemodel_b	float	Offset of custom pre-gamma curve
pregamma_en	bool	Pre-gamma transform enable
preweight	float	Weight of previous frame’s motion information
range_h	float	Horizontal motion detection window radius (7 = 15×15)
range_v	float	Vertical motion detection window radius (7 = 15×15)
sadweight	float	Weight for motion difference type A
thr_motion_slope	float	Transition slope; above this = 100% motion region

Tuning steps: 1. Enable tnr, nlm, dilate, erode, pregamma. 2. Set filter_len for reference frame damping. 3. Set filter_len2 for motion frame damping. 4. Use noise_level to separate foreground and background. 5. Adjust thr_motion_slope, sad_weight, and preweight to define the motion region.

When TNR is enabled: sensor/ISP Hblank ≥ 180 pixel clocks (256 recommended); Vblank minimum ≥ 60 sensor ET lines.

---

Demosaic

Description: Converts Bayer RAW data to RGB via interpolation. Also supports denoising, sharpening, demoire, and depurple.

Parameter	Type / Range	Description
demosaic_enable	bool	Demosaic enable
demosaic_thr	int, 0–255	Interpolation threshold for R/B channels; below = non-directional interpolation
dmsc_dir_thr_min	int, 0–4095	G channel interpolation threshold in dark areas
dmsc_dir_thr_max	int, 0–4095	G channel interpolation threshold in bright areas
dmsc_denoise_strength	int, 0–31	Low-frequency denoising strength
dmsc_sharpen_enable	bool	Sharpen enable
dmsc_sharpen_clip_black	int, 0–2047	Dark edge sharpening clip limit
dmsc_sharpen_clip_white	int, 0–2047	Bright edge sharpening clip limit
dmsc_sharpen_factor_black	int, 0–511	Dark edge sharpening factor; higher = stronger
dmsc_sharpen_factor_white	int, 0–511	Bright edge sharpening factor; higher = stronger
dmsc_sharpen_line_enable	bool	Line sharpening enable
dmsc_sharpen_line_r1	int, 0–255	Sharpening strength for values below threshold
dmsc_sharpen_line_r2	int, 0–255	Sharpening strength for values above (thr + (1<<thr_shift1))
dmsc_sharpen_line_strength	int, 0–4095	Line sharpening strength; higher = stronger
dmsc_sharpen_line_thr	int	Line sharpening threshold
dmsc_sharpen_line_thr_shift1	int, 0–10	Line sharpening threshold shift parameter
dmsc_sharpen_r1	int, 0–256	Sharpening curve parameter
dmsc_sharpen_r2	int, 0–256	Sharpening curve parameter
dmsc_sharpen_r3	int, 0–256	Sharpening curve parameter
dmsc_sharpen_size	int, 0–16	High-frequency signal representation; smaller = more fine detail sharpened
dmsc_sharpen_t1	int, 0–2047	Sharpening curve value point t1
dmsc_sharpen_t2_shift	int, 0–11	Sharpening curve value point (t1 + (1<<t2_shift))
dmsc_sharpen_t3	int, 0–2047	Sharpening curve value point t3
dmsc_sharpen_t4_shift	int, 0–11	Sharpening curve value point (t3 + (1<<t4_shift))
dmsc_demoire_enable	bool	Demoire enable
dmsc_demoire_area_thr	int, 0–32	Moire area threshold; only regions above this are processed
dmsc_demoire_r1	int, 0–255	Moire removal strength for values below t1
dmsc_demoire_r2	int, 0–255	Moire removal strength for values above (t1 + (1<<t2_shift))
dmsc_demoire_t1	int, 0–255	Moire removal strength curve value point t1
dmsc_demoire_t2_shift	int, 0–8	Moire removal strength curve value point (t1 + (1<<t2_shift))
demoire_edge_r1	int, 0–255	Moire transition curve rate for values below t1
demoire_edge_r2	int, 0–255	Moire transition curve rate for values above (t1 + (1<<t2_shift))
demoire_edge_t1	int, 0–511	Moire transition curve value point t1
demoire_edge_t2_shift	int, 0–9	Moire transition curve value point (t1 + (1<<t2_shift))
dmsc_demoire_sat_shrink	int, 0–32	Saturation reduction in moire regions; larger = stronger
dmsc_depurple_enable	bool	Purple fringe (depurple) enable
dmsc_depurple_cbcr_mode	int, 0–3	0=off, 1=R channel, 2=B channel, 3=R+B
dmsc_depurple_sat_shrink	int, 0–8	Saturation reduction in purple fringe areas
dmsc_depurple_thr	int, 8–255	Purple fringe detection threshold; smaller = more pixels treated as purple

Tuning notes:

• Interpolation: for 12-bit RAW, dir_thr_min applies at pixel value 0 and dir_thr_max at 4095; intermediate values are interpolated. demosaic_thr: gradient diff > threshold → directional interpolation; < threshold → average interpolation. Smaller values cause more false color in high-frequency regions.

• Sharpening/Denoising: denoise_strength=0 disables denoising in demosaic (usually preferred; enable only for very noisy images). A steeper sharpening curve = stronger sharpening. sharpen_line smooths horizontal/vertical lines when noise is high. The basic effect of sharpen_line is to smooth lines or textures in the horizontal or vertical direction. Larger sharpen_line_r1 and sharpen_line_r2 values make the sharpening result closer to directional (horizontal or vertical) enhancement; however, more points may be incorrectly connected, which can introduce errors in high-frequency details. factor > clip should be avoided — it over-sharpens without enough tone gradation.

• Demoire: moire correction uses a transition curve and a removal strength curve; see referenced diagrams.

---

ManualWB

Description: Manually set white balance gain values.

Parameter	Type / Range	Description
driver_load	bool	Whether to load parameters from this module
gain	float[4], 1.0–3.999	White balance gain for all four channels

---

CCM (Color Correction Matrix)

Description: Linear color space correction using a standard 3×3 matrix and vector offset. Transforms sensor color space to sRGB.

Parameter	Type / Range	Description
enable	bool	CCM enable
ccmatrix	float[9], -8.0–7.996	3×3 color correction matrix
ccoffset	int[3], -2048–2047	Per-channel offset

---

Gamma

Description: Non-linear brightness transformation to match the characteristics of typical output devices.

Parameter	Type / Range	Description
enable	bool	Gamma enable
standard	bool	Standard gamma enable
standard_val	float	Gamma value (default 2.2)
curve	int[64]	64-point gamma curve

Tuning note: Default is gamma 2.2. Custom gamma curves can be defined as needed.

---

EE (Edge Enhancement)

Description: Sharpens image detail and texture. Also suppresses black/white fringing at edges.

Parameter	Type / Range	Description
enable	bool	EE enable
ee_strength	int, 0–128	EE overall strength
ee_src_strength	int, 0–255	Denoising strength; larger = stronger. Default 1
ee_y_up_gain	int, 0–10000	Bright edge gain
ee_y_down_gain	int, 0–10000	Dark edge gain
ee_uv_gain	int, 0–1024	Edge color saturation control; larger = more saturation reduction
ee_edge_gain	int, 0–10000	Edge detail detection strength; larger = more edges detected

---

CA (Color Adjustment)

Description: Adjusts image saturation using a UV gain curve. Primarily eliminates color noise in dark or low-saturation regions. Also reduces false colors in overexposed highlights and white balance deviation in low-saturation areas.

Parameter	Type / Range	Description
ca_en	bool	CA enable
curve_en	bool	ca_curve and dci_curve enable
ca_mode	int, 0–2	0=adjust saturation by luma; 1=adjust by original saturation; 2=both
ca_curve	float	64-point CA curve

---

DCI (Dynamic Contrast Improve)

Description: Adjusts global image contrast.

Parameter	Type / Range	Description
dci_en	bool	DCI enable
dci_curve	float	64-point DCI curve

---

CProcess (Color Processing)

Description: Color processing in the YUV domain.

Parameter	Type / Range	Description
enable	bool	CProcess enable
luma_in	int	Luminance input range: 0=[64..940], 1=full [0..1023]
luma_out	int	Luminance output clipping: 0=[16..235], 1=[0..255]
chroma_out	int	Chrominance output clipping: 0=[16..240], 1=full [0..255]
bright	float, -128–127	Brightness adjustment
contrast	float, 0.3–1.9921875	Contrast adjustment
hue	float, -90–89	Hue adjustment
saturation	float, 0–1.9921875	Saturation adjustment

---

Compand

Description: Data compression and expansion module.

Parameter	Type / Range	Description
enable	bool	Compression/expansion enable
compress_enable	bool	Compression enable
compress_curve_x	int[64], 0–31	Compression X-axis interval curve
compress_use_out_y_curve	bool	Use output Y-axis curve for compression
compress_curve_y	int[64], 0–16777216	Compression Y-axis value curve
expand_enable	bool	Expansion enable
expand_curve_x	int[64], 0–31	Expansion X-axis interval curve
expand_use_out_y_curve	bool	Use output Y-axis curve for expansion
expand_curve_y	int[64], 0–16777216	Expansion Y-axis value curve

---

CAC (Chromatic Aberration Correction)

Description: Corrects lateral chromatic aberration.

Parameter	Type / Range	Description
a_blue	float, -16–15.9375	Radial blue shift parameter (formula: a_blue·r + b_blue·r² + c_blue·r³)
a_red	float, -16–15.9375	Radial red shift parameter (formula: a_red·r + b_red·r² + c_red·r³)
b_blue	float, -16–15.9375	Radial blue shift parameter
b_red	float, -16–15.9375	Radial red shift parameter
c_blue	float, -16–15.9375	Radial blue shift parameter
c_red	float, -16–15.9375	Radial red shift parameter
cac_enable	bool	CAC enable
center_h_offs	int	Horizontal distance between image center and optical center
center_v_offs	int	Vertical distance between image center and optical center

a/b/c_blue → XML blue_parameters; a/b/c_red → XML red_parameters; center_h_offs → XML x_offset; center_v_offs → XML y_offset.