nx_salience.nx

nx_salience.nx -- Itti-Koch multi-scale salience (V8). Salience = "where does the eye go first" before any semantic understanding. Itti+Koch 1998 model: compute feature pyramids (intensity, color, orientation), take center-surround differences across scales, normalize, sum. Locally distinctive regions emerge. We ship a simplified pure-math variant: - intensity pyramid (grayscale via Rec.709) - edge-magnitude pyramid (Sobel) - center-surround differences at 3 scale ratios - sum into single 8-bit saliency map What this unlocks: + automatic subject detection ("where's the main thing") + cropping suggestions (frame the high-salience region) + image-quality (boring images have flat salience) + auto-focus / auto-exposure region suggestion Per CAPTAIN_MORONI_DOCTRINE this measures IMAGE distinctiveness. It does not classify subjects; it does not rank persons; it only identifies regions that stand out from their surround. genealogy_id: itti_koch_1998_salience + koch_ullman_1985_attention lineage_id: center_surround_difference + multiscale_integration axioms: NX_AX_NORMALIZATION_PRESERVES_RANK