Box Filtering

Box filtering computes the mean or sum of pixel values within a rectangular neighborhood. The boxFilter function handles normalization, while sqrBoxFilter processes squared pixel values.

#include <opencv2/opencv.hpp>
#include <iostream>
using namespace cv;
using namespace std;

int main() {
    Mat srcImg = imread("input.jpg", IMREAD_COLOR);
    if (srcImg.empty()) {
        cerr << "Error loading image" << endl;
        return -1;
    }
    imshow("Original", srcImg);

    float sampleVals[25] = {1,2,3,4,5,
                            6,7,8,9,10,
                            11,12,13,14,15,
                            16,17,18,19,20,
                            21,22,23,24,25};
    Mat sampleMat(5, 5, CV_32FC1, sampleVals);

    Mat floatImg;
    srcImg.convertTo(floatImg, CV_32F, 1.0/255.0);

    Mat normResult, rawResult, sqrNorm, sqrRaw, floatSqrs;
    boxFilter(srcImg, rawResult, -1, Size(3,3), Point(-1,-1), false);
    boxFilter(srcImg, normResult, -1, Size(3,3), Point(-1,-1), true);
    sqrBoxFilter(sampleMat, sqrNorm, -1, Size(3,3), Point(-1,-1), true);
    sqrBoxFilter(sampleMat, sqrRaw, -1, Size(3,3), Point(-1,-1), false);
    sqrBoxFilter(floatImg, floatSqrs, -1, Size(3,3), Point(-1,-1), true);

    imshow("Normalized", normResult);
    imshow("Summation", rawResult);
    imshow("Squared Norm", floatSqrs);

    waitKey(0);
    destroyAllWindows();
    return 0;
}

Separable Filters

Separable filters decompose 2D convolution into sequential 1D operations along orthogonal axes, improving efficiency. The sepFilter2D function impelments this approach.

#include <opencv2/opencv.hpp>
#include <iostream>
using namespace cv;
using namespace std;

int main() {
    float sampleVals[25] = {1,2,3,4,5,
                            6,7,8,9,10,
                            11,12,13,14,15,
                            16,17,18,19,20,
                            21,22,23,24,25};
    Mat sampleMat(5, 5, CV_32FC1, sampleVals);

    Mat vertKernel = (Mat_<float>(3,1) << -1, 3, -1);
    Mat horizKernel = vertKernel.t();
    Mat combinedKernel = vertKernel * horizKernel;

    Mat gaussX = getGaussianKernel(3, 1);
    Mat gaussDirect, gaussSep;
    GaussianBlur(sampleMat, gaussDirect, Size(3,3), 1);
    sepFilter2D(sampleMat, gaussSep, -1, gaussX, gaussX);

    cout << "Gaussian Direct:\n" << gaussDirect << endl;
    cout << "Gaussian Separated:\n" << gaussSep << endl;

    Mat tempVert, tempComb, directConv, sepConv;
    filter2D(sampleMat, tempVert, -1, vertKernel);
    filter2D(tempVert, tempComb, -1, horizKernel);
    filter2D(sampleMat, directConv, -1, combinedKernel);
    sepFilter2D(sampleMat, sepConv, -1, horizKernel, vertKernel);

    cout << "Vertical Pass:\n" << tempVert << endl;
    cout << "Combined Pass:\n" << tempComb << endl;
    cout << "Direct 2D:\n" << directConv << endl;
    cout << "Separated 1D:\n" << sepConv << endl;

    Mat img = imread("input.jpg");
    if (img.empty()) {
        cerr << "Error loading image" << endl;
        return -1;
    }
    imshow("Source", img);

    Mat imgVert, imgComb, imgDirect;
    filter2D(img, imgVert, -1, vertKernel);
    filter2D(imgVert, imgComb, -1, horizKernel);
    filter2D(img, imgDirect, -1, combinedKernel);

    imshow("Vertical Stage", imgVert);
    imshow("Combined Result", imgComb);
    imshow("Direct 2D Conv", imgDirect);

    waitKey(0);
    destroyAllWindows();
    return 0;
}