nx_lab_from_rgb.nx

nx_lab_from_rgb.nx -- sRGB u8 -> CIELab Q10 in pure i64 fixed-point. CAPABILITY_COMPLETENESS: FULL Bridges the "RGB-proxy" tier-1 chroma axes (in nx_image_grade_v0) up to genuine colorimetric measurements: - real ITA degrees (via nx_skin_tone_ita) - real CIELab h_ab undertone angle (via nx_undertone_class) Math pipeline (CIE 15:2004 / CIE 014:2018): step 1 sRGB u8 channel value [0, 255] -> Q14 sRGB scalar [0, 16384] step 2 sRGB Q14 -> linear-light Q14 via sRGB OETF inverse: val_q14 < 0.04045 * Q14 -> linear = val / 12.92 val_q14 >= 0.04045 * Q14 -> linear = ((val + 0.055)/1.055)^2.4 Composed against nx_aces_srgb_to_linear_q14 (already shipped). step 3 Linear sRGB Q14 -> XYZ Q14 via 3x3 matrix (D65): [0.4124564 0.3575761 0.1804375] per IEC 61966-2-1 + [0.2126729 0.7151522 0.0721750] ITU-R BT.709 primaries [0.0193339 0.1191920 0.9503041] step 4 XYZ Q14 -> Lab Q10: tn_x = X / Xn (Xn = 0.95047 for D65) tn_y = Y / Yn (Yn = 1.00000) tn_z = Z / Zn (Zn = 1.08883) f(t) = t^(1/3) if t > 216/24389 (kappa*t + 16) / 116 otherwise (kappa = 24389/27) L* = 116 * f(tn_y) - 16 a* = 500 * (f(tn_x) - f(tn_y)) b* = 200 * (f(tn_y) - f(tn_z)) Output: L*_q10, a*_q10, b*_q10 all in Q10 scaled units. L*_q10 in [0, ~100 * 1024 = 102400] a*_q10 / b*_q10 in approximately [-130 * 1024, +130 * 1024] Cube root via Newton-Raphson in Q14 (5 iterations -> ~0.1% accuracy across [0, 1]).