Foreground Segmentation Improvement by Image Denoising Preprocessing Applied to Noisy Video Sequences

This work deals with the problem of recognizing the relevant pixels in a noisy video sequence (foreground segmentation). Instead of creating a new complex noise-resistant foreground segmentation method, the video frames are preprocessed to remove noise before inputting them to simpler and faster existing foreground segmentation methods. Autoencoder neural networks based on dense and convolutional layers as well as classical methods such as Mean and Median filters are selected as denoising methods in the preprocessing phase. The aim of the experiments is twofold: to verify whether old methods working on preprocessed images can be as effective as recent methods in the state of the art working on non-preprocessed images, and to study whether autoencoders can be adequate for noise removal.