如何用C++（OpenCV）移除像素和小于阈值的图像连通组件？

Got it, let's walk through how to replicate your Python OpenCV functionality in C++ step by step. This implementation will mirror your workflow: binarization, connected component detection, and removing components that fall below your pixel sum (or area) threshold.

In Python you used cv2.connectedComponents(), but in C++ we'll use cv::connectedComponentsWithStats—this function gives us both the label map and critical statistics (like component area or pixel sum) for each connected region, which is perfect for filtering.

Below is a full, commented code example. I'll cover two scenarios: one where you want to filter by component area (equivalent to pixel count in a binary image) and another where you filter by the sum of pixel values (for grayscale/color images).

Scenario 1: Filter by Component Area (Binary Image Pixel Count) #

This matches your original Python use case where "pixel sum" refers to the number of foreground pixels in the binary image:

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

using namespace cv;
using namespace std;

int main() {
    // Load input image (grayscale for simplicity)
    Mat src = imread("input_image.png", IMREAD_GRAYSCALE);
    if (src.empty()) {
        cerr << "Oops, couldn't load the image—check the file path!" << endl;
        return -1;
    }

    // Step 1: Binarize the image (using Otsu's auto-threshold; adjust as needed)
    Mat binary;
    threshold(src, binary, 0, 255, THRESH_BINARY | THRESH_OTSU);

    // Step 2: Get connected components with stats (8-connectivity, same as Python's connectivity=8)
    Mat labels, stats, centroids;
    int num_components = connectedComponentsWithStats(binary, labels, stats, centroids, 8, CV_32S);

    // Set your threshold (e.g., remove components smaller than 10 pixels)
    const int area_threshold = 10;

    // Step 3: Create output image and remove small components
    Mat output = src.clone();
    for (int i = 1; i < num_components; ++i) {
        // Extract the area of the current component
        int component_area = stats.at<int>(i, CC_STAT_AREA);
        if (component_area < area_threshold) {
            // Set all pixels in this component to background (black, adjust if needed)
            output.setTo(0, labels == i);
        }
    }

    // Visualize results
    imshow("Original Image", src);
    imshow("Binary Image", binary);
    imshow("Output (Small Components Removed)", output);
    waitKey(0);

    // Save the result
    imwrite("output_image.png", output);

    return 0;
}

Scenario 2: Filter by Sum of Pixel Values (Grayscale/Color Images) #

If "pixel sum" refers to the total of all pixel values in the component (not just count), use this approach to calculate the sum explicitly:

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

using namespace cv;
using namespace std;

int main() {
    Mat src = imread("input_image.png", IMREAD_GRAYSCALE);
    if (src.empty()) {
        cerr << "Failed to load image!" << endl;
        return -1;
    }

    // Binarize (or skip if you're working directly with grayscale)
    Mat binary;
    threshold(src, binary, 0, 255, THRESH_BINARY | THRESH_OTSU);

    // Get connected components
    Mat labels, stats, centroids;
    int num_components = connectedComponentsWithStats(binary, labels, stats, centroids, 8, CV_32S);

    // Calculate pixel sum for each component
    vector<int> component_pixel_sum(num_components, 0);
    for (int y = 0; y < labels.rows; ++y) {
        for (int x = 0; x < labels.cols; ++x) {
            int label = labels.at<int>(y, x);
            if (label != 0) { // Skip background
                component_pixel_sum[label] += src.at<uchar>(y, x);
            }
        }
    }

    // Set your sum threshold (e.g., 10)
    const int sum_threshold = 10;

    // Remove components below the threshold
    Mat output = src.clone();
    for (int i = 1; i < num_components; ++i) {
        if (component_pixel_sum[i] < sum_threshold) {
            output.setTo(0, labels == i);
        }
    }

    // Display and save
    imshow("Original", src);
    imshow("Output", output);
    waitKey(0);
    imwrite("output_sum_filtered.png", output);

    return 0;
}

• Connectivity: The 8 parameter in connectedComponentsWithStats ensures we use 8-connectivity, just like your Python example.
• Stats Matrix: The stats matrix stores data like component area (CC_STAT_AREA), bounding box coordinates, etc.—you can explore other fields if needed for more advanced filtering.
• Color Image Adjustments: If working with color images, modify the pixel sum calculation to iterate over each channel (e.g., src.at<Vec3b>(y,x)[0] for blue) and sum values across channels.

内容的提问来源于stack exchange，提问作者Saania