A TWO-STAGE UNDERWATER ENHANCEMENT NETWORK BASED ON STRUCTURE DECOMPOSITION AND CHARACTERISTICS OF UNDERWATER IMAGING

Computer Engineering

The underwater image typically has low contrast, colour distortion, and fuzzy features because of the light's attenuation and scattering in the water. By taking into account the specifics of underwater photography, a unique two-stage underwater image convolutional neural network (CNN) based on structure decomposition for underwater picture enhancement is offered as a solution to these issues. On the basis of a theoretical analysis of the underwater imaging, the raw underwater image is specifically divided into high-frequency and low-frequency components. On the foundation of a simultaneous estimation of illumination and reflectance in the linear domain, a new probabilistic method for picture enhancement is proposed. We demonstrate that in comparison to logarithmic domain, the linear domain A prototype may describe more precisely prior knowledge for better assessment of reflectance and illumination. It uses a maximum a posteriori (MAP) formulation with light and reflectance priors. The MAP problem is solved by using factors in sequential order to calculate estimates light and reflectance. The experimental results demonstrate the good performance of the suggested approach to obtain lighting and reflectance with increased visual results and a promising convergence rate. Proposed testing approach produces comparable or superior results on both subjective and objective evaluations when compared to previous testing methods. The efficiency of each component is confirmed by the ablation study, and application experiments further demonstrate how different methods can provide underwater photographs with improved visual quality.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.