// Aplica adelgazamiento a imgBinaria
void Proxy::adelgazar(bool desenfocar) {
// Nos aseguramos de que va a aparecer la imagen de fondo
if (imgPincel.empty()) {
imgPincel = imgBinaria.clone();
provider->imgPincel = imgPincel;
}
// recortamos la imagen para aumentar eficiencia
Rect mbr = MBR(imgBinaria);
qDebug() << "MBR: " << mbr.x << ", " << mbr.y << " - " << mbr.width << " x " << mbr.height;
Mat recorte = imgBinaria(mbr);
// desenfocamos y adelgazamos
if (desenfocar) {
applyGaussianBlur(recorte, 9, 0, 0);
}
execThinningGuoHall(recorte);
provider->imgBinaria = imgBinaria;
}
cv::Scalar SSIM::computeSSIM(const cv::Mat& img1, const cv::Mat& img2)
{
int ht = img1.rows;
int wt = img1.cols;
int w = wt - 10;
int h = ht - 10;
cv::Mat mu1(h,w,CV_32F), mu2(h,w,CV_32F);
cv::Mat mu1_sq(h,w,CV_32F), mu2_sq(h,w,CV_32F), mu1_mu2(h,w,CV_32F);
cv::Mat img1_sq(ht,wt,CV_32F), img2_sq(ht,wt,CV_32F), img1_img2(ht,wt,CV_32F);
cv::Mat sigma1_sq(h,w,CV_32F), sigma2_sq(h,w,CV_32F), sigma12(h,w,CV_32F);
cv::Mat tmp1(h,w,CV_32F), tmp2(h,w,CV_32F), tmp3(h,w,CV_32F);
cv::Mat ssim_map(h,w,CV_32F), cs_map(h,w,CV_32F);
// mu1 = filter2(window, img1, 'valid');
applyGaussianBlur(img1, mu1, 11, 1.5);
// mu2 = filter2(window, img2, 'valid');
applyGaussianBlur(img2, mu2, 11, 1.5);
// mu1_sq = mu1.*mu1;
cv::multiply(mu1, mu1, mu1_sq);
// mu2_sq = mu2.*mu2;
cv::multiply(mu2, mu2, mu2_sq);
// mu1_mu2 = mu1.*mu2;
cv::multiply(mu1, mu2, mu1_mu2);
cv::multiply(img1, img1, img1_sq);
cv::multiply(img2, img2, img2_sq);
cv::multiply(img1, img2, img1_img2);
// sigma1_sq = filter2(window, img1.*img1, 'valid') - mu1_sq;
applyGaussianBlur(img1_sq, sigma1_sq, 11, 1.5);
sigma1_sq -= mu1_sq;
// sigma2_sq = filter2(window, img2.*img2, 'valid') - mu2_sq;
applyGaussianBlur(img2_sq, sigma2_sq, 11, 1.5);
sigma2_sq -= mu2_sq;
// sigma12 = filter2(window, img1.*img2, 'valid') - mu1_mu2;
applyGaussianBlur(img1_img2, sigma12, 11, 1.5);
sigma12 -= mu1_mu2;
// cs_map = (2*sigma12 + C2)./(sigma1_sq + sigma2_sq + C2);
tmp1 = 2*sigma12 + C2;
tmp2 = sigma1_sq + sigma2_sq + C2;
cv::divide(tmp1, tmp2, cs_map);
// ssim_map = ((2*mu1_mu2 + C1).*(2*sigma12 + C2))./((mu1_sq + mu2_sq + C1).*(sigma1_sq + sigma2_sq + C2));
tmp3 = 2*mu1_mu2 + C1;
cv::multiply(tmp1, tmp3, tmp1);
tmp3 = mu1_sq + mu2_sq + C1;
cv::multiply(tmp2, tmp3, tmp2);
cv::divide(tmp1, tmp2, ssim_map);
// mssim = mean2(ssim_map);
double mssim = cv::mean(ssim_map).val[0];
// mcs = mean2(cs_map);
double mcs = cv::mean(cs_map).val[0];
cv::Scalar res(mssim, mcs);
return res;
}
