Image Denoising via Multiscale Feature Extraction and Triplet Attention based CNN
Ahmet Ulu, Bekır Dızdaroğlu · 2023
Recent advances in deep learning have had a significant impact on the field of noise reduction and other image-processing tasks. One of the most important innovations of deep learning is the use of convolutional layers to extract features from images. However, most convolutional neural networks (CNNs) designed for noise reduction only perform feature extraction at a single scale. Additionally, the attention mechanism, which is commonly used for channel or spatial weighting of features, neglects the interdependence and interaction of features between different dimensions. This study proposes a new CNN network for noise reduction that addresses these limitations. The proposed network uses a triplet attention mechanism and multi-scale feature extraction to capture the interactions and interdependencies between different dimensions of features. Experimental results show that the proposed approach can effectively remove Gaussian noise from grayscale images and achieves state-of-the-art performance in both quantitative and qualitative evaluations.