Demosaicing using Dual Layer Feedforward Neural Network

Prakhar Amba (1) , David Alleysson (1) , Martial Mermillod (1)

1 LPNC - Laboratoire de Psychologie et NeuroCognition

Color demosaicing is the problem of recovering full color/spectral channel from a subsampled image captured by single-chip digital cameras covered by Color Filter Array (CFA). Several algorithms have been proposed in the literature, however most of them are tuned to a particular arrangement of color filters. In this paper, we propose a generic algorithm based on a simple Neural Network (NN) architecture, which is trained on a small image database and which gives competitive results. To prove our statement, we test our network on several state of art CFAs and a 5 channel Spectral Filter Array (SFA). We demonstrate our result both on simulated images coming from standard image databases and also RAW images for a 5 channel SFA camera .

Domaines: Intelligence artificielle [cs.AI]; Traitement du signal et de l'image [eess.SP]; Traitement des images [eess.IV]

Prakhar Amba, David Alleysson, Martial Mermillod. Demosaicing using Dual Layer Feedforward Neural Network. CIC 2018, Nov 2018, Vancouver, Canada. ⟨hal-02106659⟩