CNN with Attention-Guided Concatenation for Improved Image Restoration

The performance of Deep ConNets (DCNNs) architectures has been improved in a number of ways. However, deeper Convolutional Neural Networks (CNNs) suffer from the difficulty of being trained, because widening the network introduces additional parameters and makes the denoising model more complex. Another approach is to combine image features that have been taken from various CNN architectures. In order to boost the width and, consequently, the performance of the combined network, which is better than when the networks are utilized individually, this research aims to create an improved CNN model by combining two networks. The proposed approach uses an Attention Module (ATT) at the output of the individual CNNs to select important information from the two networks before the concatenation operation. The CNN With Attention module (CWATT) performed better than the CNN with No Attention (CNOATT) on both Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM).