Self-Adaptive Colour Calibration of Deep Underwater Images Using FNN and SfM-MVS-Generated Depth Maps

Abstract

This PhD thesis addresses the challenge of colour attenuation in deep-water underwater images captured under artificial light. Initially, the research aimed to model and solve the issue through physical and mathematical approaches. However, due to challenges such as unknown environmental parameters and light reflection angles, the focus shifted to developing a more practical solution. The key innovation of this thesis is the creation of a Self-Adaptive Colour Calibration pipeline, which leverages Structure from Motion (SfM) and Multi-View Stereo (MVS) derived data, alongside DL techniques, to restore the true colours in underwater images without requiring detailed environmental knowledge. The proposed pipeline effectively corrects colour degradation in archival datasets, in the absence of spectral and environmental data, making it particularly suitable for archaeological underwater imaging and applications with limited resources. The results from the application of this methodology to datasets from two distinct underwater sites, Mazotos and Nissia, demonstrate its robustness and adaptability across varying conditions. A reference-based evaluation further confirmed the pipeline's ability to restore colours accurately, highlighting its potential for enhancing underwater imagery for archaeological and environmental studies.