Underwater image enhancement using virtual optical lens modelling and pyramid transformer network with enhanced haze optimization

Underwater images often suffer from poor colour reproduction and low contrast due to light attenuation and scattering. Most existing enhancement methods do not adequately consider the physics of underwater image formation and perceptual optimization simultaneously. To address these limitations, this paper proposes a novel hybrid framework. Initially, the Human Eye Piece Model (HEPM) utilizes geometrical optics principles to generate a structurally enhanced intermediate image. The enhanced output is then processed using the Pyramid-Shaped Transformer Network (PSTN), where hierarchical patch-based embedding and attention mechanisms effectively capture local features and contextual dependencies, resulting in improved colour balance, contrast, and texture representation. Finally, the Enhanced Haze Optimization Algorithm (EHOA) adaptively tunes system parameters using Underwater Image Quality Measure (UIQM) as the objective function for optimization. Experimental evaluation on the EUVP Dataset achieving a UIQM score of 9.50, along with reduced distortion metrics, including BRISQUE and NIQE scores of 22.50 and 3.80, respectively.