From: Steinar H. Gunderson Date: Sun, 14 Oct 2012 01:12:32 +0000 (+0200) Subject: Find Y from the xyz and the D65 white point instead of using the luma coefficients... X-Git-Tag: 1.0~253 X-Git-Url: https://git.sesse.net/?p=movit;a=commitdiff_plain;h=4625d0a9601e33201f257c68392c4b1830388a79 Find Y from the xyz and the D65 white point instead of using the luma coefficients from the standards; for sRGB, that was right, but for Rec. 601 it certainly was not. (This was massively fiddly, but in the end became surprisingly simple.) --- diff --git a/colorspace_conversion_effect.cpp b/colorspace_conversion_effect.cpp index 2f74a27..89123d3 100644 --- a/colorspace_conversion_effect.cpp +++ b/colorspace_conversion_effect.cpp @@ -6,14 +6,14 @@ // Color coordinates from Rec. 709; sRGB uses the same primaries. double rec709_x_R = 0.640, rec709_x_G = 0.300, rec709_x_B = 0.150; double rec709_y_R = 0.330, rec709_y_G = 0.600, rec709_y_B = 0.060; -double rec709_Y_R = 0.2126, rec709_Y_G = 0.7152, rec709_Y_B = 0.0722; // Color coordinates from Rec. 601. (Separate for 525- and 625-line systems.) double rec601_525_x_R = 0.630, rec601_525_x_G = 0.310, rec601_525_x_B = 0.155; double rec601_525_y_R = 0.340, rec601_525_y_G = 0.595, rec601_525_y_B = 0.070; double rec601_625_x_R = 0.640, rec601_625_x_G = 0.290, rec601_625_x_B = 0.150; double rec601_625_y_R = 0.330, rec601_625_y_G = 0.600, rec601_625_y_B = 0.060; -double rec601_Y_R = 0.299, rec601_Y_G = 0.587, rec601_Y_B = 0.114; + +double d65_x = 0.3127, d65_y = 0.3290; ColorSpaceConversionEffect::ColorSpaceConversionEffect() : source_space(COLORSPACE_sRGB), @@ -34,38 +34,86 @@ void get_xyz_matrix(ColorSpace space, Matrix3x3 m) double x_R, x_G, x_B; double y_R, y_G, y_B; - double Y_R, Y_G, Y_B; switch (space) { case COLORSPACE_REC_709: // And sRGB. x_R = rec709_x_R; x_G = rec709_x_G; x_B = rec709_x_B; y_R = rec709_y_R; y_G = rec709_y_G; y_B = rec709_y_B; - Y_R = rec709_Y_R; Y_G = rec709_Y_G; Y_B = rec709_Y_B; break; case COLORSPACE_REC_601_525: x_R = rec601_525_x_R; x_G = rec601_525_x_G; x_B = rec601_525_x_B; y_R = rec601_525_y_R; y_G = rec601_525_y_G; y_B = rec601_525_y_B; - Y_R = rec601_Y_R; Y_G = rec601_Y_G; Y_B = rec601_Y_B; break; case COLORSPACE_REC_601_625: x_R = rec601_625_x_R; x_G = rec601_625_x_G; x_B = rec601_625_x_B; y_R = rec601_625_y_R; y_G = rec601_625_y_G; y_B = rec601_625_y_B; - Y_R = rec601_Y_R; Y_G = rec601_Y_G; Y_B = rec601_Y_B; break; default: assert(false); } - // Convert xyY -> XYZ. - double X_R, X_G, X_B; - X_R = Y_R * x_R / y_R; - X_G = Y_G * x_G / y_G; - X_B = Y_B * x_B / y_B; - - double Z_R, Z_G, Z_B; - Z_R = Y_R * (1.0f - x_R - y_R) / y_R; - Z_G = Y_G * (1.0f - x_G - y_G) / y_G; - Z_B = Y_B * (1.0f - x_B - y_B) / y_B; + // Recover z = 1 - x - y. + double z_R = 1.0 - x_R - y_R; + double z_G = 1.0 - x_G - y_G; + double z_B = 1.0 - x_B - y_B; + + // Find the XYZ coordinates of D65 (white point for both Rec. 601 and 709), + // normalized so that Y=1. + double d65_X = d65_x / d65_y; + double d65_Y = 1.0; + double d65_Z = (1.0 - d65_x - d65_y) / d65_y; + + // We have, for each primary (example is with red): + // + // X_R / (X_R + Y_R + Z_R) = x_R + // Y_R / (X_R + Y_R + Z_R) = y_R + // Z_R / (X_R + Y_R + Z_R) = z_R + // + // Some algebraic fiddling yields (unsurprisingly): + // + // X_R = (x_R / y_R) Y_R + // Z_R = (z_R / y_R) Z_R + // + // We also know that since RGB=(1,1,1) should give us the + // D65 illuminant, we must have + // + // X_R + X_G + X_B = D65_X + // Y_R + Y_G + Y_B = D65_Y + // Z_R + Z_G + Z_B = D65_Z + // + // But since we already know how to express Y and Z by + // some constant multiple of X, this reduces to + // + // k1 Y_R + k2 Y_G + k3 Y_B = D65_X + // Y_R + Y_G + Y_B = D65_Y + // k4 Y_R + k5 Y_G + k6 Y_B = D65_Z + // + // Which we can solve for (Y_R, Y_G, Y_B) by inverting a 3x3 matrix. + + Matrix3x3 temp, inverted; + temp[0] = x_R / y_R; + temp[3] = x_G / y_G; + temp[6] = x_B / y_B; + + temp[1] = 1.0; + temp[4] = 1.0; + temp[7] = 1.0; + + temp[2] = z_R / y_R; + temp[5] = z_G / y_G; + temp[8] = z_B / y_B; + + invert_3x3_matrix(temp, inverted); + float Y_R, Y_G, Y_B; + multiply_3x3_matrix_float3(inverted, d65_X, d65_Y, d65_Z, &Y_R, &Y_G, &Y_B); + + // Now convert xyY -> XYZ. + double X_R = temp[0] * Y_R; + double Z_R = temp[2] * Y_R; + double X_G = temp[3] * Y_G; + double Z_G = temp[5] * Y_G; + double X_B = temp[6] * Y_B; + double Z_B = temp[8] * Y_B; m[0] = X_R; m[3] = X_G; m[6] = X_B; m[1] = Y_R; m[4] = Y_G; m[7] = Y_B;