Image denoiser with autoencoder

I’m new to knime software. I would like to build a flow to use an autoencoder to remove noise from images. I was not able to find any suggestion or idea on knime hub.
Someone can help me?

Thanks Felice

Hi @felice60 and welcome to the community!

by autoencoder, you mean a neural network that encodes the input such that it omits the unnecessary parts and denoises that way, right? I don’t think there is an example on the Hub right now, so you’d have to piece it together yourself, I’m afraid. A challenging, but great way to learn KNIME, I suppose! There might be two starting points: Fine-tune VGG16 – KNIME Hub for how to use images with neural networks in KNIME and this topic Autoencoder. Midpoint. for an autoencoder example. A pretty advanced project, I would assume, which needs quite some knowledge of neural networks.

But do you really need a full blown, well trained neural network? Maybe you can get away with the KNIME Image Processing Extension. There are quite a lot of filters there that might do the trick just as well. If you have salt-and-pepper noise, for example, I can recommend the Median Filter. If its more large scale noise, I have good experience with CLAHE. If you want to share a little bit more of what kind of images you want to denoise, I’m happy to give more suggestions.

Kind regards,
Lukas

2 Likes