Opencv notes 20 PCA

One 、 Introduction and practical application

PCA It is mainly used to obtain the main direction of the object and reduce the dimension of the data .

PCA The main idea of is to find several important characteristics that can reflect the characteristics in a pile of dimensional data , So it can reduce the amount of calculation , Eliminate those attributes that are not particularly important from these data .

Two 、 Deduction of mathematical principles

 

 

 

3、 ... and 、 opencv Medium PCA class

PCA::PCA(InputArray data, InputArray mean, int flags, int maxComponents=0)
data           // input data （ It can be a set of contour points ）
mean           // Data zero mean , It's empty （Mat（）） It's calculated automatically 
flag           // Represents the way data is provided （0 Represents input by line ,1 It means input by column ）
maxComponents  // How many eigenvalues to keep （ By default, all are reserved ）

•   Original image , Project into a new space

Mat PCA::project(InputArray vec) const

• Conduct project Later data , Reflect the original image

Mat PCA::backProject(InputArray vec) const

The variable values are ：mean-------- The mean of the raw data

                  eigenvalues-------- The eigenvalues of the covariance matrix

                  eigenvectors-------- Eigenvector

Four 、 Code shows

opencv actual combat ——PCA Application of algorithm - Only you are strong - Blog Garden

#if  1 //  PCA Algorithm   Algorithm 
// Get the main direction of building 
double getOrientation(vector<Point>& contour, Mat& img)
{
	// structure pca data . What we're doing here is we're going to take the contour points x and y As two dimensions press into data_pts In the middle .
	Mat data_pts = Mat(contour.size(), 2, CV_64FC1);// Use mat To save the data , It's also for the back pca Processing needs 
	for (int i = 0; i < data_pts.rows; ++i)
	{
		data_pts.at<double>(i, 0) = contour[i].x;
		data_pts.at<double>(i, 1) = contour[i].y;
	}
	// perform PCA analysis 
	PCA pca_analysis(data_pts, Mat(), 0);
	// Get the most important component （ mean value ）, In this case , It corresponds to the midpoint of the contour , It's also the midpoint of the image 
	Point pos = Point(pca_analysis.mean.at<double>(0, 0), pca_analysis.mean.at<double>(0, 1));
	// Store eigenvectors and eigenvalues 
	vector<Point2d> eigen_vecs(2);
	vector<double> eigen_val(2);
	for (int i = 0; i < 2; ++i)
	{
		eigen_vecs[i] = Point2d(pca_analysis.eigenvectors.at<double>(i, 0), pca_analysis.eigenvectors.at<double>(i, 1));
		eigen_val[i] = pca_analysis.eigenvalues.at<double>(i, 0);// In the outline / Draw a small circle at the midpoint of the image 
		circle(img, pos, 3, CV_RGB(255, 0, 255), 2);
		// Calculate the straight line , Draw a line in the main direction （ Each eigenvector is multiplied by its eigenvalue and converted to the average position . There is one  0.02  The scaling factor of , It's just to make sure the vector fits the image and doesn't  10000  The length of the pixel ）
		line(img, pos, pos + 0.02 * Point(eigen_vecs[0].x * eigen_val[0], eigen_vecs[0].y * eigen_val[0]), CV_RGB(255, 255, 0));
		line(img, pos, pos + 0.02 * Point(eigen_vecs[1].x * eigen_val[1], eigen_vecs[1].y * eigen_val[1]), CV_RGB(0, 255, 255));
		// Finally calculate and return the strongest ( It has the largest eigenvalue ） The angle of the eigenvector of 
		return atan2(eigen_vecs[0].y, eigen_vecs[0].x);
	}
}
int main(int argc, char** argv)
{

	Mat src = imread("C:\\Users\\19473\\Desktop\\opencv_images\\600.png");
	imshow(" The input image ", src);
	Mat gray, binary;
	cvtColor(src, gray, COLOR_BGR2GRAY);
	// Threshold processing 
	threshold(gray, binary, 150, 255, THRESH_BINARY);
	imshow(" Two valued ", binary);
	//1 、 Look for the outline 
	vector<vector<Point> > contours;
	vector<Vec4i> hierarchy;
	findContours(binary, contours, hierarchy, RETR_LIST, CHAIN_APPROX_NONE);
	// Profile analysis , Find the artifact 
	for (size_t i = 0; i < contours.size(); ++i)
	{
		// Calculate the contour size 
		double area = contourArea(contours[i]);
		// Remove small or large contour areas （ Scientific notation means le2 Express 1X10 Of 2 Power ）
		if (area < 1e2 || 1e4 < area) continue;
		// Draw the outline 
		drawContours(src, contours, i, Scalar(0, 0, 255), 2, 8, hierarchy, 0);
		// Look for the direction of each profile 
		double angle = getOrientation(contours[i], src);
		cout << angle << endl;
	}

	imshow(" result ", src);
	waitKey(0);
	return 0;
}

#endif