Lorenzo Terna
Recovering audiovisual heritage using synchrotron X-rays and machine learning techniques.
Rel. Marco Scianna. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Matematica, 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (6MB) | Preview |
Abstract
This thesis develops a unified computational pipeline to recover audio from degraded magnetic tapes using the Play It Again non-contact readout approach based on X-ray magnetic circular dichroism (XMCD). XMCD acquisitions produce sequences of high-dimensional detector frames in which the magnetization-dependent signal is weak and embedded in strong background and noise. The first contribution is an automated mask-estimation method that selects information-bearing pixels on the detector and enables a robust reduction from 2D dichroic frames to a 1D waveform. The mask is learned from a reference dichroic acquisition via unsupervised clustering on physics-inspired pixel features (intensity, local statistics, and gradient-based smoothness), with an optional translation (registration) step to compensate for small footprint drifts.
Both binary and confidence-weighted masks are investigated, showing how soft weighting can further suppress background-dominated regions
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Ente in cotutela
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
