Reservoir Computing Networks (RCNs) have already been used in different areas of artificial intelligence (e.g., robotics and brain machine interfacing). Recently, we demonstrated that RCNs are also effective for large vocabulary continuous speech recognition in clean condition and for small vocabulary continuous speech recognition in the presence of background noise. Moreover, we conceived a simple strategy for tuning an RCN to the classification problem at hand. In this work, we demonstrate that (1) the tuning strategy that was conceived for speech processing is also applicable to image processing, (2) RCN-based systems can offer state-of-the-art handwritten digit recognition, both in the absence and presence of noise and (3) RCN-based systems can denoise images and achieve good noise robust recognition by supplying these images to a recognizer that was solely trained on clean images. Our comparative experiments demonstrate that the proposed RCN-based handwritten digit recognizer achieves an error rate of 0.81% on the clean test data of the well-known MNIST benchmark and that the proposed RCN-based denoiser can reduce the error rate on the noisy test data from 37% to 2%.