Hybrid Image Denoising using DnCNN, FNet, and Wavelet Transforms

This project implements advanced image denoising models based on DnCNN, with enhancements using Wavelet Transforms and FFT-based Attention (FNet). It explores hybrid architectures for improving denoising performance on noisy images.

Model Architectures

1. DnCNN (Baseline)

• Deep convolutional network with residual learning.
• Learns to predict noise from corrupted images.

2. DnCNN + Wavelet

• Adds Discrete Wavelet Transform (DWT) preprocessing.
• Enhances noise separation by working in frequency domain.

3. HybridTriBranchModel

• Combines three branches:
  • Standard DnCNN
  • DWT + DnCNN
  • FNet (FFT-based self-attention)
• Merges outputs for robust denoising under various noise types.