Technical Report v1.20.1: A Ground-Truth-Free Binary Classifier for SSIM Structural Masking Risk, Validated Across Four Models

Technical Report v1.20 found that a single, untrained threshold on mu_notta alone classifies Amplification Phase risk at 96-100% accuracy across four models, with recall at or above 99.5% throughout, using no ground truth (GT) at application time, and left open whether an analogous GT-free signal exists for this series' other named phenomenon, Structural Masking. This report finds one: comp_contrast, one of this series' eight independently-measured quality features, is RMS contrast computed on the raw 8-bit pixel scale rather than the [0,1]-normalized scale used for sigma_gt/sigma_notta throughout this series -- dividing by 255 recovers sigma_gt from comp_contrast_gt exactly, with zero error across all four models' full sigma_gt<=0.16 populations (81,564,041 blocks each). By the same construction, comp_contrast_notta/255 is an exact, algebra-free measurement of sigma_notta -- not an approximation. A single untrained threshold, sigma_notta<=sigma_stat (the same calibrated value already established for sigma_gt), classifies Structural Masking risk at 98.83-99.53% accuracy across all four models, with masking-region recall at 99.09-99.95%. Recall on the minority non-masking class is lower and model-dependent (42.79-78.76%): a substantial share of blocks with real ground-truth texture (sigma_gt>sigma_stat) are nonetheless reconstructed with less texture than sigma_stat (sigma_notta<=sigma_stat). Because sigma_notta is measured exactly here, this gap reflects real reconstruction behavior, not measurement error. As with v1.20, this shifts when GT is needed -- from every image checked to once per model, to calibrate sigma_stat -- rather than eliminating it. Every finding is computed against the same four independent, no-join, from-scratch verification databases (Technical Report v1.12), read-only, screen-recorded and hash-verified. The complete verification SQL, covering all four models sequentially, is included to enable independent reproduction. Dataset and Mandatory Citation: - Source: NIH ChestX-ray8 (Hospital-scale chest x-ray database) - Citation: Wang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2017). "ChestX-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thoracic diseases." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 3462–3471. - Download: https://nihcc.app.box.com/v/ChestXray-NIHCC Contact: [email protected] (individual) /[email protected] (Meritcs) Published by Meritcs (https://merit-metrics.org), an independentresearch organization studying image-quality metric reliability.