// // ImageComparer.cpp // // Created by Nissim Hadar on 18 Nov 2017. // Copyright 2013 High Fidelity, Inc. // // Distributed under the Apache License, Version 2.0. // See the accompanying file LICENSE or http://www.apache.org/licenses/LICENSE-2.0.html // #include "ImageComparer.h" #include "common.h" #include // Computes SSIM - see https://en.wikipedia.org/wiki/Structural_similarity // The value is computed for the luminance component and the average value is returned double ImageComparer::compareImages(QImage resultImage, QImage expectedImage) const { // Make sure the image is 8 bits per colour QImage::Format format = expectedImage.format(); if (format != QImage::Format::Format_ARGB32) { throw -1; } const int L = 255; // (2^number of bits per pixel) - 1 const double K1 { 0.01 }; const double K2 { 0.03 }; const double c1 = pow((K1 * L), 2); const double c2 = pow((K2 * L), 2); // First go over all full 8x8 blocks // This is done in 3 loops // 1) Read the pixels into a linear array (an optimization) // 2) Calculate mean // 3) Calculate variance and covariance // // p - pixel in expected image // q - pixel in result image // const int WIN_SIZE = 8; int x{ 0 }; // column index (start of block) int y{ 0 }; // row index (start of block // Pixels are processed in square blocks double p[WIN_SIZE * WIN_SIZE]; double q[WIN_SIZE * WIN_SIZE]; int windowCounter{ 0 }; double ssim{ 0.0 }; while (x < expectedImage.width()) { int lastX = x + WIN_SIZE - 1; if (lastX > expectedImage.width() - 1) { x -= (lastX - expectedImage.width()); } while (y < expectedImage.height()) { int lastY = y + WIN_SIZE - 1; if (lastY > expectedImage.height() - 1) { y -= (lastY - expectedImage.height()); } // Collect pixels into linear arrays int i{ 0 }; for (int xx = 0; xx < WIN_SIZE; ++xx) { for (int yy = 0; yy < WIN_SIZE; ++yy) { // Get pixels QRgb pixelP = expectedImage.pixel(QPoint(x + xx, y + yy)); QRgb pixelQ = resultImage.pixel(QPoint(x + xx, y + yy)); // Convert to luminance p[i] = R_Y * qRed(pixelP) + G_Y * qGreen(pixelP) + B_Y * qBlue(pixelP); q[i] = R_Y * qRed(pixelQ) + G_Y * qGreen(pixelQ) + B_Y * qBlue(pixelQ); ++i; } } // Calculate mean double mP{ 0.0 }; // average value of expected pixel double mQ{ 0.0 }; // average value of result pixel for (int j = 0; j < WIN_SIZE * WIN_SIZE; ++j) { mP += p[j]; mQ += q[j]; } mP /= (WIN_SIZE * WIN_SIZE); mQ /= (WIN_SIZE * WIN_SIZE); // Calculate variance and covariance double sigsqP{ 0.0 }; double sigsqQ{ 0.0 }; double sigPQ{ 0.0 }; for (int j = 0; j < WIN_SIZE * WIN_SIZE; ++j) { sigsqP += pow((p[j] - mP), 2); sigsqQ += pow((q[j] - mQ), 2); sigPQ += (p[j] - mP) * (q[j] - mQ); } sigsqP /= (WIN_SIZE * WIN_SIZE); sigsqQ /= (WIN_SIZE * WIN_SIZE); sigPQ /= (WIN_SIZE * WIN_SIZE); double numerator = (2.0 * mP * mQ + c1) * (2.0 * sigPQ + c2); double denominator = (mP * mP + mQ * mQ + c1) * (sigsqP + sigsqQ + c2); ssim += numerator / denominator; ++windowCounter; y += WIN_SIZE; } x += WIN_SIZE; y = 0; } return ssim / windowCounter; };