Efficient recurrent real video restoration
Year: 2024
Type: article
Abstract: We propose a novel method dealing with the most common limitations of real video sequences, namely noise, blur, flicker and low contrast. The method is based on an intensive use of the DCT for both the deblurring and denoising tasks, combined with classical strategies for tonal stabilization and low-light enhancement. To the best of our knowledge, this is the first time that all of these problems have been addressed using a unified framework. Comparisons with state of the art learning based methods for denoising and deblurring show that the proposed approach obtains better results, with the additional advantages of being fully interpretable, having a reduced memory footprint and lighter computational requirements, which make it well suited to be incorporated into the processing chain of mobile devices.
Source: Digital Signal Processing
Authors: Antoni Buades, Jose-Luis Lisani
Institution: Universitat de les Illes Balears
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,500
Award: G4040262502
OpenAlex ID: W4404311718