Image Denoising Using Multi-Level Wavelet Convolutional Neural Network
Ameya Gawande, Shripad S. Bhatlawande, Ameya Pangavhane, Swati Shilaskar · 2024
This paper presents a technique called the Multi-Scale Wavelet Convolutional Neural Network which is developed to tackle noises that are present in an image. The method combines convolutional neural networks and wavelet transforms, used to denoise an image while maintaining important features of it. The method leverages deep learning and Discrete Wavelet Transform for comprehensive multi-scale image analysis. The SIDD dataset is used for this task which consists of noisy/ground-truth image pairs taken by 5 different smartphones under different lighting conditions. MWCNN significantly enhanced the image quality.when tested with real-world noises present in images, The observed Peak Signal to Noise Ratio of 35.42 dB, an average Structural Similarity Index of 0.8650, and an average Mean Squared Error of 22.41 for denoised images. The potential of MWCNN as an image-de noising solution is useful in medical imaging, surveillance footage enhancement, artifact removal, healthcare and security.