Image denoising is one of the fundamental problems in image processing eld since it is required by many computer vision applications. Various approaches have been used in image denoising throughout the years from spatial ltering to model based approaches. Having outperformed all traditional methods, neural network based discriminative methods have gained popularity in the recent years. However, most of these methods still struggle to achieve exibility against various noise levels and types. In this thesis, we propose a deep convolutional autoencoder combined with a variant of feature pyramid network for image denoising. We use simulated data in Blender software along with corrupted natural images during training to improve robustness against various noise levels and types. Our experimental results show that proposed method can achieve competitive performance in blind Gaussian denoising with significantly less training time required compared to state-of-the-art methods. Extensive experiments showed us proposed method gives promising performance in wide range of noise levels with a single network.