Denoise in images with attention-based convolution neural networks

The need for image quality enhancement studies has increased due to the importance of clear and realistic images for the performance of computer vision applications and for human viewing pleasure. Most image quality enhancement studies have focused on the problem of noise removal. This problem can be caused by various sources such as image acquisition, compression, and transmission, and various approaches have been tried for a long time to overcome this problem. This study is based on Toward a Fast and Flexible Solution for CNN based Image Denoising (FFDNet), one of the successful denoising methods, in which a Convolutional Block Attention Module (CBAM) has been integrated into the architecture to improve the denoising capability of FFDNet. The CBAM module contains two separate attention mechanisms, and its inclusion in the FFDNet architecture has increased the representational performance of convolutional neural networks. Ablation studies were performed with the FFDNet architecture, and it was found that the CBAM module gives successful results only when it is added to the last layer. Experiments with the proposed method were performed with different noise levels, and improvements in the PSNR, SSIM, and LRI metrics were obtained in the quantitative measurements.When analyzing the visual quality of the images, it can be seen that the proposed method provides results that are closer to the target images.