Boundary-Transition Characterization in High-Speed Underwater Bubble Imaging Under Broadband and Narrow-Band Illumination

Characterizing gas–liquid boundaries in underwater high-speed images is complicated by optical propagation and interface effects that broaden intensity transitions. Whether narrower transitions also indicate more accurate boundary localization, however, remains unclear. We integrated broadband white and narrow-band blue, green, and yellow source conditions with configuration-specific RBW processing at nominal observation distances of 0.7 and 5.0 m. Each source-distance condition was represented by one acquisition run, from which 20 temporally separated frames were analyzed as frame-level subsamples. In the 5.0 m run, the observed mean RBW was 17.46% under the blue-source condition and 36.18% under the white-source condition, corresponding to a descriptive reduction of 51.73%. A sensitivity analysis based on the interquartile range (IQR) yielded a descriptive reduction of 33.42%, whereas the near-field means were similar across source conditions. Target-plane optical irradiance and camera spectral responsivity were not calibrated. These values therefore describe source-condition responses of the complete imaging chain rather than isolated wavelength effects. A 12-image near-field holdout set annotated by two observers was used to compare the proposed estimator with classical and learning-based edge detectors. This image-reference benchmark did not support a claim of superior localization, and RBW showed little monotonic association with normalized image-reference localization discrepancy within the holdout set. The results characterize within-run image responses rather than reproducible source-condition effects. They support RBW as a descriptor of processed transition width, but not as a measure of physical boundary-localization accuracy.