FreqPatchNet: A Dual-Domain Patch-Wise Fusion Network for Robust Phase Correction in Underwater Image Reconstruction

This paper presents FreqPatchNet, a novel patch-wise dual-domain Convolutional Neural Network (CNN) designed to correct phase distortions in underwater images. The model uses bispectral frequency features and local CNN regression to reconstruct clean images from distorted inputs. Evaluated using Peak Signal-to-Noise Ratio (PSNR) and Mean Squared Error (MSE), FreqPatchNet achieves a maximum PSNR of 35.6 dB and a lowest MSE of 0.28 at 10% distortion. A comparative analysis with state-of-the-art methods shows the superior performance of the proposed model in structural similarity. Real-world tests confirm its potential for underwater robotics and vision applications.