I. Overview
Traditional image noise removal is mainly based on various filters, but they are not data-specific, so many image details may be lost, or the effect of noise removal is not ideal. Neural network-based autoencoders can be used to learn noise removal filters for datasets.
For an introduction to autoencoders, you can refer to the link below.
https://skydance.blog.csdn.net/article/details/123567960 https://skydance.blog.csdn.net/article/details/123567960 A typical autoencoder
It consists of an encoder, which converts some high-dimensional input into a low-dimensional format, and a decoder, which can read the encoded state and convert it to other formats. Encoded states are also known as latent states.
<