Is there a proper way for the Segmentation of Rocks? - opencv

Question

I work with the Segmentation of Rocks and have been using the Image Segmentation with Watershed Algorithm from Opencv(https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_watershed/py_watershed.html).
As some background on my input, I get a depth Image as Input image and need to perform segmentation on it.
I have tried to change the kernels for the Laplacian.
I have made a few tweaks to the code. cv::adaptivethreshhold seems to show better than cv::threshold in my case.
Even changing the kernels doesn't improve the result. Hence, I have left the kernels unchanged.
I also referred to questions relating to my probelm :
-> Image Sharpening Using Laplacian Filter
-> How to perform segmentation on complex images like rocks (in c++)
const cv::Mat image = rrlib::coviroa::AccessImageAsMat(*in_depth_img);
cv::Mat depth_img_opencv;
cv::extractChannel(image, depth_img_opencv,0);
cv::Mat dst;
cv::resize(depth_img_opencv,dst, cv::Size(image.cols*3,image.rows*3));
cv::imshow("Depth Image", dst);
cv::waitKey(200);
cv::Mat src;
cv::cvtColor(depth_img_opencv, src,cv::COLOR_GRAY2BGR);
cv:: Mat dst1;
cv::resize(src, dst1, cv::Size(image.cols*3,image.rows*3));
cv::imshow("Color Image", dst1);
cv::waitKey(200);
cv::Mat kernel = (cv::Mat_<float>(3,3) <<
1, 1, 1,
1, -8, 1,
1, 1, 1);
cv::Mat imgLaplacian;
cv::Mat sharp = src;
cv::filter2D(sharp, imgLaplacian, CV_32F, kernel);
src.convertTo(sharp, CV_32F);
cv::Mat imgResult = sharp - imgLaplacian;
imgResult.convertTo(imgResult, CV_8UC3);
src = imgResult;
cv:: Mat dst2;
cv::resize(src, dst2, cv::Size(image.cols*3,image.rows*3));
cv::imshow("Sharpened Image", dst2);
cv::waitKey(200);
cv::Mat bw;
cv::cvtColor(src, bw, CV_BGR2GRAY);
cv::imshow("BGR2GRAY", bw);
cv::waitKey(200);
cv::adaptiveThreshold(bw, bw, 255, CV_ADAPTIVE_THRESH_GAUSSIAN_C , CV_THRESH_BINARY, 7, 0);
cv:: Mat dst3;
cv::resize(bw, dst3, cv::Size(image.cols*3,image.rows*3));
cv::imshow("Binary Image", dst3);
cv::waitKey(200);
cv::Mat dist;
cv::distanceTransform(bw, dist, CV_DIST_L2, 3);
cv::normalize(dist, dist, 0, 1., cv::NORM_MINMAX);
cv:: Mat dst4;
cv::resize(dist, dst4, cv::Size(image.cols*3,image.rows*3));
cv::imshow("Distance Transform", dst4);
cv::waitKey(200);
// Threshold to obtain the peaks
dist.convertTo(dist, CV_8UC1);
cv::adaptiveThreshold(dist, dist, 255, CV_ADAPTIVE_THRESH_GAUSSIAN_C, CV_THRESH_BINARY, 7, 0);
// Dilate a bit the dist image
cv::Mat kernel1 = cv::Mat::ones(3, 3, CV_8UC1);
// Thickens obects in an image
cv::dilate(dist, dist, kernel1);
cv:: Mat dst5;
cv::resize(dist, dst5, cv::Size(image.cols*3,image.rows*3));
cv::imshow("Peaks", dst5);
cv::waitKey(200);
Any valuable suggestion is appreciated to get a better segmentation of the result.