The field of image manipulation is dynamic, exploiting a range of algorithms to analyze, manipulate and enhance digital images. Our study focuses on a crucial application of image processing, which is the elimination of blind Gaussian noise in order to improve image quality and facilitate image analysis by preserving essential details. In this research, we explore the use of different convolutional neural network (CNN) architectures to tackle the problem of blind Gaussian noise, applying different noise levels, ranging from low to high. We present an in-depth comparative analysis of the three main CNN architectures: DnCNN, DRNet and RIDNet, highlighting the quantitative and qualitative experimental results of these different approaches. These methods have demonstrated remarkable performance in image processing tasks, particularly denoising, using various techniques built into CNNs, such as batch normalization and residual learning. Our results show that these techniques bring significant improvements to all three CNN approaches, as evidenced by the remarkable performance observed in the experimental results. These findings underline the robustness of CNN architectures in the face of complex noise scenarios, such as the blind noise scenario addressed in our study.