Self-adaptive colour calibration pipeline for deep underwater image restoration: a reference-based evaluation

Building upon our previous work on underwater (UW) image colour restoration, this paper advances the self-adaptive colour calibration (SACC)pipeline, by introducing a systematic, reference-based quantitative evaluation. UWimagery is notoriously affected by wavelength-dependent light absorption, scattering, and colour attenuation; these factors distort the visual appearance of submerged objects and complicate scientific documentation.  Traditional model-based correction techniques often rely on well-calibrated equipment or prior environmental data; many artificial intelligence (AI)-driven approaches suffer from dataset bias and limited generalisability. The work explained below addresses these challenges by providing a flexible pipeline that restores colours without requiring extensive environmental metadata; this methodology is particularly well-suited for archival and in-field datasets. New datasets acquired from the 18th-century Nissia shipwreck site, located off Protaras in Cyprus, were employed to evaluate the pipeline under real-world excavation conditions. These datasets, captured with a rigid camera–strobe configuration, include reference regions with colour charts; these provide reliable ‘ground-truth’(GT)values for objective comparison.  Structure from Motion (SfM) and Multi-View Stereo (MVS) were used to derive depth maps, which, together with red-green-blue (RGB)values, were input to a Feedforward Neural Network (FNN). The network was trained using Adam and RMSprop optimisers, enabling the restoration of perceptually accurate colours across varying conditions. Three metrics, modified R², Euclidean Distance, and the D metric, were applied to assess performance. The evaluation is constrained by the limited number of images containing colour charts; even so, this limitation reflects authentic field conditions, where divers must prioritise site documentation over calibration. Overall, this study validates the robustness of the SACC pipeline and introduces a replicable framework for evaluating UW image colour restoration. Beyond methodological advances, the findings highlight the potential of the pipeline to support cultural heritage documentation, archaeological interpretation, and long-term preservation of submerged sites.