"OpenCV Development Notes (XXIII): basis of morphological filtering algorithm - open operation (after the first expansion of corrosion)": To be published

"OpenCV Development Notes (24): the basis of morphological filtering algorithm - closing operation (after the first expansion of corrosion)": To be published

"OpenCV Development Notes (XXV): basis of morphological filtering algorithm - white hat": to be released

"OpenCV Development Notes (26): the basis of morphological filtering algorithm - Black Hat": to be released

Continued to supplement ...

OpenCV Development Notes (XVI): The basic algorithm for linear filtering - Gaussian filter

Foreword

This chapter explains the linear filtering with the Gaussian filter.

Demo

Gaussian filter

Outline

Gaussian filtering is input to each pixel array is convolved with a Gaussian kernel acid, and a convolution output as a pixel value. After the image is smoothed Gaussian filtering process depends on the standard deviation.

Gaussian filter output is a weighted average of the art items, while the closer the pixel from the center the higher the weight, and therefore, with respect to the mean filter, it is more gentle smoothing and edge preservation was also better.

In image processing, a Gaussian filter is generally implemented in two ways:

Sliding window discrete convolution window: the discrete window when very large amount of calculation with a sliding window is very large (i.e., using separable filters implemented), the implementation may be considered based on Fourier transform .
Fourier transform;

Feature

Gaussian filtering is the most useful filter filter characteristics are as follows:

Less distortion : single-valued function is a Gaussian function, Gaussian field pixel weighted mean value of the store instead of the pixel, the pixel will change with the weight monotonically distance, to reduce distortion.
Rotational symmetry: Gaussian function having a resistance against rotation, in the respective directions Gaussian smoothing Chengdu is the same, it is difficult to estimate the noise present in the directivity, ensure smooth performance is not biased in any direction.
Not affect the frequency: the Fourier transform of the Gaussian function is a single lobe of the spectrum, a Gaussian filter is the smoothed image will not be affected by the unwanted high frequency signals, while retaining most of the desired signal.
Controllable noise parameters: Gaussian smoothing filter by the degree of the variance [sigma] determined, [sigma] , the more wide band, the better the degree of smoothing, the image noise may be provided with a controllable parameter.
Separability: a two-dimensional Gaussian convolution can be performed in two steps, the first image and the one-dimensional Gaussian function convolution, the convolution result and a direction perpendicular to the same one-dimensional Gaussian function convolution.

Function prototype

void GaussianBlur( InputArray src, 
                OutputArray dst,
                Size ksize, 
                double sigmaX,
                double sigmaY = 0,
                int borderType = BORDER_DEFAULT );

A parameter : InputArray type generally cv :: Mat, and can handle any number of image channels. Note, however, the depth of the picture should be treated as CV_8U , CV_16U , CV_16S , CV_32F , CV_64F one.
Parameter Two ; OutputArray type, output target image, you need the original picture and have the same size and type.
Three parameters : Size type ksize, the size of the sub-accounts. General use Size (w, h) to indicate the size of the sub-accounting, Size (3,3) on a 3x3 size represented by the sub-accounting, w and h may be different sizes.
Four parameters : double type sigmaX, represents the standard deviation of the Gaussian kernel function of the X-direction.
Five parameters : double type sigmaY, represents the standard deviation of the Gaussian kernel function of the Y-direction.

Above, when sigmaX and sigmaY are 0, the default calculated using sub-accounts:

Six parameters : int type borderType, inferring mode for a certain image of the external boundary pixels, the default values BORDER_DEFULAT, generally without use.

Source Demo

void OpenCVManager::testGaussianBlurFilter()
{
    QString fileName1 = "E:/qtProject/openCVDemo/openCVDemo/modules/openCVManager/images/1.jpg";
    cv::Mat srcMat = cv::imread(fileName1.toStdString());

    cv::String windowName = _windowTitle.toStdString();
    cvui::init(windowName);

    if(!srcMat.data)
    {
        qDebug() << __FILE__ << __LINE__
                 << "Failed to load image:" << fileName1;
        return;
    }

    cv::Mat dstMat;
    dstMat = cv::Mat::zeros(srcMat.size(), srcMat.type());
    cv::Mat windowMat = cv::Mat(cv::Size(dstMat.cols * 3, dstMat.rows),
                                srcMat.type());
    int ksize = 1;      // 核心大小
    int sigmaX = 0;  // x方向的标准偏差
    int sigmaY = 0;  // y方向的标准偏差
    cvui::window(windowMat, dstMat.cols, 0, dstMat.cols, dstMat.rows, "settings");
    while(true)
    {
        windowMat = cv::Scalar(0, 0, 0);
        // 原图先copy到左边
        cv::Mat leftMat = windowMat(cv::Range(0, srcMat.rows),
                                    cv::Range(0, srcMat.cols));
        cv::addWeighted(leftMat, 1.0f, srcMat, 1.0f, 0.0f, leftMat);
        // 中间为调整滤波参数的相关设置
        cvui::printf(windowMat, 375, 20, "ksize = size *  2 + 1");
        cvui::trackbar(windowMat, 375, 30, 165, &ksize, 0, 10);

        cvui::printf(windowMat, 375, 80, "sigmaX");
        cvui::trackbar(windowMat, 375, 90, 165, &sigmaX, 0, 100);

        cvui::printf(windowMat, 375, 140, "sigmaY");
        cvui::trackbar(windowMat, 375, 150, 165, &sigmaY, 0, 100);

        // 高斯滤波
        cv::GaussianBlur(srcMat, dstMat, cv::Size(ksize * 2 + 1, ksize * 2 + 1), sigmaX / 10.f, sigmaY / 10.f);

        // 效果图copy到右边
        // 注意：rang从位置1到位置2，不是位置1+宽度
        cv::Mat rightMat = windowMat(cv::Range(0, srcMat.rows),
                                     cv::Range(srcMat.cols * 2, srcMat.cols * 3));
        cv::addWeighted(rightMat, 0.0f, dstMat, 1.0f, 0.0f, rightMat);
        // 更新
        cvui::update();
        // 显示
        cv::imshow(windowName, windowMat);
        // esc键退出
        if(cv::waitKey(25) == 27)
        {
            break;
        }
    }
}