VisionX: A Hybrid Deep Learning Framework for Image Restoration

Single-image deraining is a challenging low-level computer vision task due to the varying distribution, density, and spatial patterns of rain streaks, which obscure scene details and degrade image quality. This paper presents VisionX, a hybrid deep learning framework for image restoration that integrates wavelet-based feature extraction with transformer-based contextual modeling. The proposed architecture employs the Haar Discrete Wavelet Transform (DWT) to decompose images into low- and high-frequency components, a Restormer encoder–decoder for spatial restoration, and Swin Transformer blocks for hierarchical feature extraction and contextual modeling. Multi-scale feature fusion, progressive refinement, and reconstruction, perceptual, and frequency-aware loss components are incorporated to improve rain-streak suppression while preserving background structures and fine image details. The framework is designed for single-image deraining and is evaluated using image restoration metrics, including Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), on benchmark datasets. The study explores the potential of combining spatial-domain transformer modeling with frequency-domain processing for effective image restoration. Keywords: Image Deraining, VisionX, Restormer, Swin Transformer, Haar Wavelet Transform, Deep Learning, Image Restoration.