nx_quant_table.nx

nx_quant_table.nx -- perceptual quantisation matrices. The two 64-entry tables from ITU-T Recommendation T.81 (JPEG) Annex K -- fully public since 1992 -- give a 30-50% rate-distortion advantage over uniform quantisation at moderate quality on natural images. Math: each post-DCT coefficient at zig-zag position (r,c) is divided by qt[r*8+c], where qt is the quant matrix. Higher divisors for high-frequency coefficients aggressively discard the perceptually-unimportant detail; small divisors for the low-frequency corner preserve structure. Quality factor Q in 1..100 maps to a scale factor: Q >= 50: scale = 200 - 2 * Q (Q=50 -> 100; Q=100 -> 0) Q < 50: scale = 5000 / Q (Q=1 -> 5000; Q=49 -> 102) Then qt[i] = clamp((base[i] * scale + 50) / 100, 1, 255). genealogy_id: itu_t_t_81_1992_annex_k + jpeg_iso_10918_1992 + nx_dct8_q10 lineage_id: nishi_quant_table_q10