Iris Fractal & Nevi Analysis: Comparative Study of Pigment Architecture and Pathology Markers — Argira Station (v5.1)

Abstract

Updated release v5.1. Robustness fixes applied after validation on real biometric images (Kaggle MMU NIR dataset) and synthetic simulator. Architecture unchanged — fixes improve segmentation stability across iris phenotypes and image modalities. SUBJECT A (brown iris, calibrated): 29 nevi. D_global=1.868.D_nevi=1.572±0.115. Auric Index A=0.847. p=0.993. SUBJECT B/OL (green-hazel, sectorial heterochromia, calibrated): 23 nevi. D_global=1.836. D_nevi=1.629±0.086. Auric Index A=0.840. p=0.999. SYNTHETIC VALIDATION (brown_dark, golden_org=0.78 injected):Auric Index A=0.847 recovered. p=0.963. D_global=1.849.Confirms Auric Index correctly measures golden angle organization. EXTERNAL VALIDATION (Kaggle MMU NIR, 2 images):Pipeline runs end-to-end without errors on real external images.n_nevi=1-3 (expected for NIR biometric format, not clinical). NEW v5.1 fixes:- NEVI_MIN_AREA: 30 → 80 px (reduces edge fragmentation)- Adaptive LAB sigma: 1.5 → 1.0 (NIR + visible color compatible)- Saturation filter percentile: 25% → 10% (permissive for NIR)- QC checks: fatal errors → non-fatal warnings (analysis always continues)- Morphological opening 3×3 added in adaptive LAB segmentation Both subjects maintain A≈0.84 with p>0.99.Validation requires n=40-50 cohort. Related preprint: https://doi.org/10.5281/zenodo.18781344Phyllotaxis base: https://doi.org/10.5281/zenodo.18760369Dataset v2: https://doi.org/10.5281/zenodo.18799576Dataset v3: https://doi.org/10.5281/zenodo.18803498

Ähnliche Arbeiten