Underwater image enhancement using opencv c++

Question

I am trying to enhance an underwater video image using opencv. The object detection is taking place in the HSV color space. Prior to that, I have been trying to figure out the technique to eliminate the color distortion from the water. One technique I read about was to contrast stretch the RGB color space and then in HSI stretch the saturation and intensity.

In a attempt to produce something similar that works I came up with using normalize on BGR and then converting to HSV and normalizing the saturation and value. This doesn't seem to do the trick of eliminating the blue. Is there something wrong with my order or have I missed something in underwater image enhancement?

while(1){
//store image to matrix
capture.read(cameraFeed);
feedClone = cameraFeed.clone();

Mat HSV;
vector<Mat> channels;
vector<Mat> hsv_planes;

/*This is the part I am hoping to get feedback on*/

split(cameraFeed,channels);
normalize(channels[0], channels[0], 0, 255, NORM_MINMAX);
normalize(channels[1], channels[1], 0, 255, NORM_MINMAX);
normalize(channels[2], channels[2], 0, 255, NORM_MINMAX);
merge(channels,cameraFeed);     

cvtColor(cameraFeed,HSV,COLOR_BGR2HSV);
hsv_planes.clear();
split(HSV,hsv_planes);
normalize(hsv_planes[1], hsv_planes[1], 0, 255, NORM_MINMAX);
normalize(hsv_planes[2], hsv_planes[2], 0, 255, NORM_MINMAX);
merge(hsv_planes,HSV);
cvtColor(HSV,cameraFeed,COLOR_HSV2BGR);


/*This is what happens next and works perfectly out of water without the above adjustments*/

//This finds the specific color in the threshold         
cvtColor(cameraFeed,HSV,COLOR_BGR2HSV);
inRange(HSV,orange.getHSVmin(),orange.getHSVmax(),threshold);

//this function runs the threshold through 2 erodes and 2 dilates
//then a median blur (7,7)
morphOps(threshold);

//this tracks that image in the feed
trackFilteredObject(orange,threshold,HSV,feedClone);
}

Answer 1

Look at this Github repository which has gathered a collection of underwater image restoration and enhancements in Matlab scripts. It covers different solutions such as color space handling, convolutional neural networks (CNN), medium transmission, pyramids and structured edge detectors. You can also run them with Octave.

It also provides a top script to run these methods altogether. The results of each solution is saved on disk allowing quality evaluation of each method. You can see the results and choose one of them. The only thing you need to do is to write its C++ equivalent. There are also some Python underwater image restoration and enhancements in this repository. The Python codes are much more difficult to be written in C++.

Answer 2

Try this code. I had the same problem which is solved by it to great extent.

#include "opencv2/opencv.hpp"
#include <iostream>

using namespace std;
using namespace cv;

int main(int argc, char** argv)
{

    cout<<"Usage: ./executable input_image output_image \n";

    if(argc!=3)
    {
        return 0;
    }


    int filterFactor = 1;
    Mat my_img = imread(argv[1]);
    Mat orig_img = my_img.clone();
    imshow("original",my_img);

    Mat simg;

    cvtColor(my_img, simg, CV_BGR2GRAY);

    long int N = simg.rows*simg.cols;

    int histo_b[256];
    int histo_g[256];
    int histo_r[256];

    for(int i=0; i<256; i++){
        histo_b[i] = 0;
        histo_g[i] = 0;
        histo_r[i] = 0;
    }
    Vec3b intensity;

    for(int i=0; i<simg.rows; i++){
        for(int j=0; j<simg.cols; j++){
            intensity = my_img.at<Vec3b>(i,j);

            histo_b[intensity.val[0]] = histo_b[intensity.val[0]] + 1;
            histo_g[intensity.val[1]] = histo_g[intensity.val[1]] + 1;
            histo_r[intensity.val[2]] = histo_r[intensity.val[2]] + 1;
        }
    }

    for(int i = 1; i<256; i++){
        histo_b[i] = histo_b[i] + filterFactor * histo_b[i-1];
        histo_g[i] = histo_g[i] + filterFactor * histo_g[i-1];
        histo_r[i] = histo_r[i] + filterFactor * histo_r[i-1];
    }

    int vmin_b=0;
    int vmin_g=0;
    int vmin_r=0;
    int s1 = 3;
    int s2 = 3;

    while(histo_b[vmin_b+1] <= N*s1/100){
        vmin_b = vmin_b +1;
    }
    while(histo_g[vmin_g+1] <= N*s1/100){
        vmin_g = vmin_g +1;
    }
    while(histo_r[vmin_r+1] <= N*s1/100){
        vmin_r = vmin_r +1;
    }

    int vmax_b = 255-1;
    int vmax_g = 255-1;
    int vmax_r = 255-1;

    while(histo_b[vmax_b-1]>(N-((N/100)*s2)))
    {   
        vmax_b = vmax_b-1;
    }
    if(vmax_b < 255-1){
        vmax_b = vmax_b+1;
    }
    while(histo_g[vmax_g-1]>(N-((N/100)*s2)))
    {   
        vmax_g = vmax_g-1;
    }
    if(vmax_g < 255-1){
        vmax_g = vmax_g+1;
    }
    while(histo_r[vmax_r-1]>(N-((N/100)*s2)))
    {   
        vmax_r = vmax_r-1;
    }
    if(vmax_r < 255-1){
        vmax_r = vmax_r+1;
    }

    for(int i=0; i<simg.rows; i++)
    {
        for(int j=0; j<simg.cols; j++)
        {

            intensity = my_img.at<Vec3b>(i,j);

            if(intensity.val[0]<vmin_b){
                intensity.val[0] = vmin_b;
            }
            if(intensity.val[0]>vmax_b){
                intensity.val[0]=vmax_b;
            }


            if(intensity.val[1]<vmin_g){
                intensity.val[1] = vmin_g;
            }
            if(intensity.val[1]>vmax_g){
                intensity.val[1]=vmax_g;
            }


            if(intensity.val[2]<vmin_r){
                intensity.val[2] = vmin_r;
            }
            if(intensity.val[2]>vmax_r){
                intensity.val[2]=vmax_r;
            }

            my_img.at<Vec3b>(i,j) = intensity;
        }
    }

    for(int i=0; i<simg.rows; i++){
        for(int j=0; j<simg.cols; j++){

            intensity = my_img.at<Vec3b>(i,j);
            intensity.val[0] = (intensity.val[0] - vmin_b)*255/(vmax_b-vmin_b);
            intensity.val[1] = (intensity.val[1] - vmin_g)*255/(vmax_g-vmin_g);
            intensity.val[2] = (intensity.val[2] - vmin_r)*255/(vmax_r-vmin_r);
            my_img.at<Vec3b>(i,j) = intensity;
        }
    }   


    // sharpen image using "unsharp mask" algorithm
    Mat blurred; double sigma = 1, threshold = 5, amount = 1;
    GaussianBlur(my_img, blurred, Size(), sigma, sigma);
    Mat lowContrastMask = abs(my_img - blurred) < threshold;
    Mat sharpened = my_img*(1+amount) + blurred*(-amount);
    my_img.copyTo(sharpened, lowContrastMask);    

    imshow("New Image",sharpened);
    waitKey(0);

    Mat comp_img;
    hconcat(orig_img, sharpened, comp_img);
    imwrite(argv[2], comp_img);
}

Check here for more details.

Answer 3

Have u tried to read this link? http://answers.opencv.org/question/75510/how-to-make-auto-adjustmentsbrightness-and-contrast-for-image-android-opencv-image-correction/

void Utils::BrightnessAndContrastAuto(const cv::Mat &src, cv::Mat &dst, float clipHistPercent)
{

    CV_Assert(clipHistPercent >= 0);
    CV_Assert((src.type() == CV_8UC1) || (src.type() == CV_8UC3) || (src.type() == CV_8UC4));

    int histSize = 256;
    float alpha, beta;
    double minGray = 0, maxGray = 0;

    //to calculate grayscale histogram
    cv::Mat gray;
    if (src.type() == CV_8UC1) gray = src;
    else if (src.type() == CV_8UC3) cvtColor(src, gray, CV_BGR2GRAY);
    else if (src.type() == CV_8UC4) cvtColor(src, gray, CV_BGRA2GRAY);
    if (clipHistPercent == 0)
    {
        // keep full available range
        cv::minMaxLoc(gray, &minGray, &maxGray);
    }
    else
    {
        cv::Mat hist; //the grayscale histogram

        float range[] = { 0, 256 };
        const float* histRange = { range };
        bool uniform = true;
        bool accumulate = false;
        calcHist(&gray, 1, 0, cv::Mat(), hist, 1, &histSize, &histRange, uniform, accumulate);

        // calculate cumulative distribution from the histogram
        std::vector<float> accumulator(histSize);
        accumulator[0] = hist.at<float>(0);
        for (int i = 1; i < histSize; i++)
        {
            accumulator[i] = accumulator[i - 1] + hist.at<float>(i);
        }

        // locate points that cuts at required value
        float max = accumulator.back();
        clipHistPercent *= (max / 100.0); //make percent as absolute
        clipHistPercent /= 2.0; // left and right wings
        // locate left cut
        minGray = 0;
        while (accumulator[minGray] < clipHistPercent)
            minGray++;

        // locate right cut
        maxGray = histSize - 1;
        while (accumulator[maxGray] >= (max - clipHistPercent))
            maxGray--;
    }

    // current range
    float inputRange = maxGray - minGray;

    alpha = (histSize - 1) / inputRange;   // alpha expands current range to histsize range
    beta = -minGray * alpha;             // beta shifts current range so that minGray will go to 0

    // Apply brightness and contrast normalization
    // convertTo operates with saurate_cast
    src.convertTo(dst, -1, alpha, beta);

    // restore alpha channel from source 
    if (dst.type() == CV_8UC4)
    {
        int from_to[] = { 3, 3 };
        cv::mixChannels(&src, 4, &dst, 1, from_to, 1);
    }
    return;
}