ISP pipeline hdrplus

Denoise,HDR,Isppipeline,Image-processing(图形处理),camera, Isp ,HDRplus

817
209
C

分享一个全网性能最快的raw转yuv图处理sdk,一键测试raw视频(普通pc 手机能实时60fp 1080p)

链接:https://pan.baidu.com/s/1n3FtWCY1Rn2y8gUlKDXkHQ 
提取码:aaaa

QQ技术讨论资料群236898700

详解v1.0 v3.0 测试使用以及各个模块详细教程qq [email protected]

《编译测试教程链接》

链接:https://pan.baidu.com/s/19nbjRKWprGrEFpVKtpzsXg
提取码:aaaa

Mat

1,Introduce

debug
ref

Matlib Is an open source (C / C + +) image algorithm library, It can be used under multiple platforms(Window, Linux, Android), It contains all kinds of conventional image processing algorithms, all functions adopt the underlying optimization technology, include (Neon, Openmp, Linebuf, Mempool, and so on) ,Its performance is faster than OpenCV, The code is only 1M, which is very simplified and has no dependency,It is very convenient to use, We can use the neon instruction on the window platform, Development and use is very friendly, which can quickly integrate developers into their own code.

2,Frame structure

framestucture

Here we briefly describe the core class of mat basic image library (parent class mat.h file)
Subclasses contain (MultIntImage, MultiShortImage, MultiUcharImage, MultiUshortImage, SingleUcharImage, Yuv420Image), They contain various basic image processing algorithms, We can use them to process images such as RAW RGB GRAY YUV.
For detailed code usage, please refer to the example file.

3,How to compile a project

  • Window platform: We can use vs2017 to compile(open ISPpipeline.sln)
  • Linux platform: We can use g++ to compile(makefile)
  • Android platform: We can use ndk to compile(android.mk)

---------------------------V2.0-----------------------------

3,Example for surperlowlight(wait for…)

注意事项本算法要求是多帧ev0和多帧ev-x(详细参考demo教材)

测试图片demo下载链接:https://pan.baidu.com/s/1EWJfSL-iD62dzy0qodP7ww

提取码:aaaa

标定工具下载:https://pan.baidu.com/s/1EWJfSL-iD62dzy0qodP7ww

提取码:aaaa

基于mat surperlowlight 框架图

ISPpipeline.png

1,常规配置

comom_param.png

效果图如图:

common_isp.png

2,surperlowlight效果配置

superlowlight_param.png

效果图如图:

superlowlight_isp.png

3,细节对比

detailcompare.png

---------------------------V1.0-----------------------------

4,Example for hdrplus(https://github.com/jhfmat/ISP-pipeline-hdrplus/releases/tag/v1.0)

注意事项本算法要求是多帧短曝光raw

测试图片demo下载链接:https://pan.baidu.com/s/1AAD0IRYJ3sOZy8ENNXo4Yg

提取码:aaaa

####(DecedeCR2.exe文件将.CR2文件解码到.raw文件和生成相关的参数文件.txt,ISPpipeline.exe是一套isp将raw处理到bmp流程)
run23detail.png

基于mat hdrplus 框架图

ISPpipeline.png

1,打开网络参数文件

weight.param.png

2,关闭isp种核心三个模块

  • nBlockMatchFusionEnable=0; ValueRange=[0,1,1] 多帧去噪
  • nChromaDenoiseEnable=0; ValueRange=[0,1,1] 去除彩色噪声
  • nTonemappingEnable=0; ValueRange=[0,1,1] 动态范围压缩

如图:

sampleweight.png

3,运行run23.bat

run23.png

Outbmp目录下生成Normalize.bmp如图:

basicispss.png

4,打开网络模块开启上述三个模块如图:

fullweight.png

删除outbmp/Normalize.bmp 再次运行run23.bat

outbmp目录下生成Normalize.bmp图如下:

myispss.png

5,下载hdrplus里面的demo测试结果图项目链接

https://github.com/timothybrooks/hdr-plus

hdrplusziplianjie.png
hdrpluszip.png

6,解压001压缩包找到里面output23.png

hdrplusdemooutdir.png

hdrplus测试结果图

hdrplusout23ss.png

7,三张图整体对比

fullcompare.png

8,三张细节对比

detailcompare.png