This repo is an implementation of ISP in c function, basically it is re-implemented from openISP in python. It's currently in very first version, I list some TODO list to make pipeline more robust.
And if you find something wrong in my implementation, welcome to let me know.
Function list
-
Dead Pixel Correction
-
Black Level Compensation
-
Len Shading Correction(need calibration parameters)
-
Anti-aliasing Noise Filter
-
AWB gain Control
-
CFA interpolation
only implement rggb bayer pattern
-
Color Correction
-
Gamma Correction
piecewise LUT
-
Color Space Conversion
-
Noise Filter for chroma
- False Color Suppression
-
Hue Saturation Control
-
Noise Filter for Luma
- Non local mean denoise
- Bilateral Filter
-
Edge Enhancement
-
Contrast Brightness Control
build exe file,
mkdir build/
cd build
cmake ..
cmake --build .
then isp_pipeline.exe will in prebuilt/ folder
test isp pipeline
isp_pipeline.exe test.rawThe configuration can be a adjusted as input raw image.
In the config file , parameters are floats which are converted to fixed point type in C function to make operation in integer type.
config.json
{
"width": 1920,
"height": 1080,
"image_bits": 10,
"blc": {
"bl_r": 0,
"bl_gr": 0,
"bl_gb": 0,
"bl_b": 0,
"alpha": 0.0,
"beta": 0.0
},
"dpc": {
"dead_thres": 30,
"mode": "gradient"
},
"bayer_pattern": "rggb",
"cfa_mode": "malvar",
"awb_gain": {
"r_gain": 1.5,
"gr_gain": 1.0,
"gb_gain": 1.0,
"b_gain": 1.1
},
"ccm": [
[
1.0,
0.0,
0.0,
0.0
],
[
0.0,
1.0,
0.0,
0.0
],
[
0.0,
0.0,
1.0,
0.0
]
],
"gamma": 0.8,
"csc": [
[0.2568,0.5041,0.0979,16],
[-0.1482,-0.291,0.4392,128],
[0.4392,-0.3678,-0.0714,128]
],
"nlm":{
"Ds":4,
"ds":1,
"h":5
},
"bnf":{
"dw":[
[8,12,32,12,8],
[12,64,128,64,12],
[32,128,1024,128,32],
[12,64,128,64,12],
[8,12,32,12,8]
],
"rw":[0,8,16,32],
"rthres":[128,32,8]
},
"edge_filter":[
[-1,0,-1,0,-1],
[-1,0,8,0,-1],
[-1,0,-1,0,-1]
],
"eeh":{
"gain_min":32,
"gain_max":128,
"thres_min":4,
"thres_max":16,
"em_clip_min":-64,
"em_clip_max":64
},
"fcs":{
"edge_min":32,
"edge_max":64
},
"hsc":{
"hue_offset":0,
"saturation_gain":1.3
},
"bcc":{
"brightness":0,
"contrast":1.1
}
}
-
board condition (I didn't handle any board condition in any function)
-
lens shading correction (if I have parameters)
-
CFA ( with more bayer pattern)
-
color space conversion (with more yuv format)
-
Non local mean denoise (optimization in exponential operation)
-
bilateral filter (with non-fixed kernel and optimize in fixed point operation)
-
False color suppression (the current result is kind of weird, need to find the reference algorithm)s
-
Hue Saturation Control (the current result is kind of weird, need to find the reference algorithm)