A novel image Denoising approach using super resolution densely connected convolutional networks

Image distortion effects, called noise, may occur due to various reasons such as image acquisition, transfer, and duplication. Image denoising is a preliminary step for many studies in the field of image processing. The vast majority of techniques in the literature require parameters that the user must determine according to the noise intensity. Due to the user requirement, the developed techniques become almost impossible to use by another computer system. Therefore, the Densely Connected Convolutional Networks structure-based model is proposed to remove noise from gray-level images with different noise levels in this study. With the developed approach, the obligation of the user to enter any parameters has been eliminated. For the training of the proposed method, 2200 noisy images with 11 different levels derived from the BSDS300 Train dataset (original 200 images) were used, and the success of the method was evaluated with 1100 noisy images derived from the BSDS300 Test dataset (original 100 images). The images used to evaluate the success of the proposed method were compared to both the traditional and state-of-the-art techniques. It was observed that the average SSIM / PSNR values obtained with the proposed method for the whole test dataset were 0.9236 / 33.94 at low noise level (sigma(2) = 0.001) and 0.7156 / 26.39 at high noise level (sigma(2) = 0.020). The results show that the proposed method is a very effective and efficient noise filter for image denoising.