Noise Removal Using a Multi-Convolutional Block and Transfer Learning-Based Neural Network
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This study presents a new learning algorithm that combines convolutional neural networks (CNNs) with transfer learning to efficiently remove additive white Gaussian noise.The algorithm uses transfer learning to reduce training time by initializing with the weights of a pre-trained model and applies multiscale learning to improve noise removal while preserving the original image's detailed structure.Images for transfer learning are generated using the local steering kernel and modified steering kernel filter, which capture both local and nonlocal information and effectively preserve key structural elements, such as borders and textures.Experimental results show that the proposed algorithm outperforms traditional filtering techniques and other deep learning-based methods in restoration experiments at various noise levels on the Set12 data set.Quantitative evaluation using peak signal to noise ratio (PSNR) confirms that the algorithm achieves high PSNR values and demonstrates strong restoration performance in most noisy environments.
Fields of Study & Subjects
Image and Signal Denoising Methods
Image Enhancement Techniques
Advanced Image Processing Techniques
Publication details
- Journal / source
- Journal of information and communication convergence engineering
- Volume
- 24
- Issue
- 1
- Pages
- 59–70
- Publication date
- 31 March 2026
- Language
- EN