随着图像处理技术和计算机视觉技术的蓬勃发展，对特殊天气下的场景检测和图像处理成为重要的研究方向。在雾天拍摄的图像容易受雾或霾的影响，导致图片模糊、对比度低以至于丢失图像重要信息。因此，需要对带雾图像进行去雾，处理图像信息，保证其他计算机视觉任务的正常运行。

一、halcon算法

1.1 halcon算法源码

本算法用到的图像见资源链接，已上传至资源文件

**********************************
*何凯明博士去雾算法代码实现
*论文：<<Single Image Haze Removal Using Dark Channel Prior>>
*编写时间：2016-04-11
*作者：datiansong
**********************************
dev_update_off ()
dev_close_window ()
read_image (Image, 'fish')
get_image_size (Image, Width, Height)
dev_open_window (0, 0, Width, Height, 'black', WindowHandle)
dev_display (Image)
disp_message (WindowHandle, '原图像', 'window', 12, 12, 'red', 'false')
*转换图像类型，用于后续运算
convert_image_type (Image, IxImage, 'real')
*求取暗通道图像
decompose3 (IxImage, R, G, B)
min_image (R, G, ImageMin)
min_image (ImageMin, B, ImageMin1)
gray_erosion_rect (ImageMin1, DarkChannelImage,5, 5)
*计算全球大气光成分A的值
min_max_gray (DarkChannelImage, DarkChannelImage, 0.1, Min, Max, Range)
threshold (DarkChannelImage, Region, Max, 255)
min_max_gray (Region, IxImage, 0, Min1, A, Range1)
*计算透视率预估值tx
scale_image (IxImage, ImageScaled, 1/A, 0)
decompose3 (ImageScaled, R1, G1, B1)
min_image (R1, G1, ImageMin2)
min_image (ImageMin2, B1, ImageMin3)
*==================================================特别注意，下面的参数需要进行适当的，本人提供的图和参数直接用即可
*下面的尺寸如果是原来的15，那么楼房的边会出现涂抹的效果，很难看
gray_erosion_rect (ImageMin3, ImageMin4, 3, 3)
*下面的小数，绝对值越大，颜色越深，在这张图上，为-0.6效果相对较好，何博士的原来为-0.95很黑
scale_image (ImageMin4, txImage, -0.7, 1)
*设定阈值T0，如果t<T0,则t=T0
T0:=0.1
threshold (txImage, Region1, 0, T0)
paint_region (Region1, txImage, txImage, T0, 'fill')
*求取去雾后的图像
scale_image (IxImage, ImageScaled1, 1, -A)
decompose3 (ImageScaled1, R2, G2, B2)
div_image (R2, txImage, ImageResultR, 1, A)
div_image (G2, txImage, ImageResultG, 1, A)
div_image (B2, txImage, ImageResultB, 1, A)
compose3 (ImageResultR, ImageResultG, ImageResultB, JxImage)
dev_display (Image)
dev_open_window (0, 0+Width, Width, Height, 'black', WindowHandle1)
dev_display (JxImage)
disp_message (WindowHandle1, '去雾图', 'window', 12, 12, 'green', 'false')

1.2 halcon算法效果图

二、opencv算法

2.1 python源码

python的算法很简单，实际上仅调用了opencv的算法，一共4行主要代码就实现了功能，但是这个算法相对于其他的算法而言，可调的参数几乎没有。

import numpy as np
import cv2

if __name__ == '__main__':
    img = cv2.imread('fog1.png')

    # 实现去雾代码
    b, g, r = cv2.split(img)
    bx, gx, rx = cv2.equalizeHist(b), cv2.equalizeHist(g), cv2.equalizeHist(r)
    img_enhance = cv2.merge((bx, gx, rx))

    images = np.concatenate((img, img_enhance), axis=1)
    cv2.imwrite('fog1_enhance.jpeg', images)
    cv2.imshow('result', images)
    cv2.waitKey()
    cv2.destroyAllWindows()

2.2opencv算法效果图

三、C#算法

3.1 C#源码

using System;
using System.Collections.Generic;
using System.ComponentModel;
using System.Data;
using System.Drawing;
using System.Drawing.Imaging;
using System.Linq;
using System.Text;
using System.Windows.Forms;
using System.Runtime.InteropServices;
using System.Diagnostics;
namespace HazeRemovalTest
{
    public unsafe partial class FrmTest : Form
    {
        // dll的代码中用的是StdCall，这里也要用StdCall，如果用Cdecl，则会出现对 PInvoke 函数“....”的调用导致堆栈不对称错误，再次按F5又可以运行
       
        [DllImport("HazeRemoval.dll", CallingConvention = CallingConvention.StdCall, CharSet = CharSet.Unicode, ExactSpelling = true)]
        private static extern void HazeRemovalUseDarkChannelPrior(byte* Src, byte* Dest, int Width, int Height, int Stride, int Radius,int GuideRadius, int MaxAtom, float Omega, float Epsilon, float T0);

        private bool Busy = false;

        public FrmTest()
        {
            InitializeComponent();
        }
     

        private void CmdOpen_Click(object sender, EventArgs e)
        {
            OpenFileDialog openFileDialog = new OpenFileDialog();
            openFileDialog.RestoreDirectory = true;
            if (openFileDialog.ShowDialog() == DialogResult.OK)
            {
                PicSrc.Image.Dispose();
                PicDest.Image.Dispose();
                PicSrc.Image = Bitmap.FromFile(openFileDialog.FileName);
                PicDest.Image = Bitmap.FromFile(openFileDialog.FileName);
                Application.DoEvents();
                ShowHazeRemovalResult();
            }
        }

        private void CmdHazeRemoval_Click(object sender, EventArgs e)
        {
            ShowHazeRemovalResult();
        }

        private void ShowHazeRemovalResult()
        {
            Busy = true;
            Bitmap SrcB = (Bitmap)PicSrc.Image;
            Bitmap DstB = (Bitmap)PicDest.Image;
            BitmapData SrcBmpData = SrcB.LockBits(new Rectangle(0, 0, SrcB.Width, SrcB.Height), ImageLockMode.ReadWrite, PixelFormat.Format24bppRgb);
            BitmapData DstBmpData = DstB.LockBits(new Rectangle(0, 0, DstB.Width, DstB.Height), ImageLockMode.ReadWrite, PixelFormat.Format24bppRgb);
            Stopwatch Sw = new Stopwatch();
            Sw.Start();
            HazeRemovalUseDarkChannelPrior((byte*)SrcBmpData.Scan0, (byte*)DstBmpData.Scan0, SrcBmpData.Width, SrcBmpData.Height, SrcBmpData.Stride, BlockSize.Value, GuideBlockSize.Value, MaxAtom.Value, Omega.Value * 0.01f, Epsilon.Value * 0.001f, T0.Value * 0.01f);
            Sw.Stop();
            this.Text = Sw.ElapsedMilliseconds.ToString();

            SrcB.UnlockBits(SrcBmpData);
            DstB.UnlockBits(DstBmpData);
            PicDest.Invalidate();
            Busy = false;
        }

        private void FrmTest_Load(object sender, EventArgs e)
        {
            ShowHazeRemovalResult();
        }

        private void BlockSize_Scroll(object sender, ScrollEventArgs e)
        {
            LblBlockSize.Text = BlockSize.Value.ToString();
            if (Busy==false) ShowHazeRemovalResult();
        }

        private void GuideBlockSize_Scroll(object sender, ScrollEventArgs e)
        {
            LblGuideBlockSize.Text = GuideBlockSize.Value.ToString();
            if (Busy == false) ShowHazeRemovalResult();
        }

        private void Omega_Scroll(object sender, ScrollEventArgs e)
        {
            LblOmega.Text = Omega.Value.ToString() + "%";
            if (Busy == false) ShowHazeRemovalResult();
        }

        private void MaxAtom_Scroll(object sender, ScrollEventArgs e)
        {
            LbLAtom.Text = MaxAtom.Value.ToString();
            if (Busy == false) ShowHazeRemovalResult();
        }

        private void Epsilon_Scroll(object sender, ScrollEventArgs e)
        {
            LblEpsilon.Text = (Epsilon.Value * 0.0001).ToString();
            if (Busy == false) ShowHazeRemovalResult();
        }

        private void T0_Scroll(object sender, ScrollEventArgs e)
        {
            LblT0.Text = (T0.Value * 0.01).ToString();
            if (Busy == false) ShowHazeRemovalResult();
        }

  
    }
}

下载地址

http://files.cnblogs.com/Imageshop/HazeRemovalTest.rar

四、VB源码

4.1 截图

下载地址

http://files.cnblogs.com/Imageshop/%E5%9B%BE%E5%83%8F%E5%8E%BB%E9%9B%BE%E7%BB%BC%E5%90%88%E7%89%88%E6%9C%AC.rar

五、matlab源码

下载地址

https://link.csdn.net/?target=http%3A%2F%2Ffiles.cnblogs.com%2FImageshop%2Fcvpr09defog%2528matlab%2529.rar

说明：本文所需图片以及python源码已放置本人的资源中，自行下载

https://download.csdn.net/download/sunnyrainflower/87952490?spm=1001.2014.3001.5503

六、总结

关于去雾的算法，本人更喜欢C#的代码，可调参数多，且容易实现，python的程序几乎没有可调参数；halcon的算法还需要精调参数，和图像的尺寸和清晰度比较相关；功能最多的是vb的代码，源码的作者集成了6种算法在里面，都可以尝试一下。matlab的就不做过多的说明了，自己慢慢研究吧。

更多的除雾算法及论文说明
参考链接
https://blog.csdn.net/huixingshao/article/details/42834939
https://zhuanlan.zhihu.com/p/489222309