2D-filter design based on Volterra equation and machine learning to tackle the aliasing effect in image demosicing
Year: 2025
Type: article
Abstract: For nonlinear problems, the discrete Volterra equation is a helpful model. An aliasing effect between components from a CFA image is known as the main problem in Demosaicing on frequency-domain analysis, which denotes a nonlinear problem. The discrete Volterra equa-tion can be used as a model to solve Demosaicing. A suitable Volterra kernel is required. We show that there is an analogy between the weights and bias of the perceptron and the coefficients of the Volterra kernel. This analogy allows us to discover the coefficients of 2D-nonlinear filters by numerically solving the discrete Volterra equation. On the images from the Kodak database, this Demosaicing procedure is applied in accordance with the Volterra equation's order and the Volterra kernel's size. Effectiveness is judged using several measures. The study's findings show that the aliasing effect is lessened depending on the order of the Volterra equation and the Volterra kernel size chosen. Our mean signal-to-noise ratio performance, using a 3x3 kernel size and the second-order Volterra equation, is 37.7 dB.
DOI: 10.14419/50xc1e97
Authors: Nodjigoto Blague, Boukar Ousman, Libouga Li Gwet David, Kamgang Jean Claude, Ndogotar Nelio
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4408689252