Demosaicking aims to approximate missing color pixels through analysis of the geometric structure between given color pixels and missing color pixels. In this paper, we introduce an efficient adaptive demosaicking method based on back propagation (BP) neural network (BP-NN). We firstly reconstruct the green channel using one BPNN, and then refine the green channel utilizing another BPNN based on the color difference. With the whole green channel interpolated, we reconstruct the red/blue channel using the color difference between the green channel and red/blue channel in a local region. Finally, we refine the red/blue channel using the third BPNN. Regarding the interpolation issue, different image features have completely different properties, such as smooth regions, edges, and textures. Consequently, it is necessary to identify an adaptive model to estimate the relation among neighboring color pixels. We provide the adaptive BP-NN based demosaicking algorithm which can reduce blurring through recovery of missing pixels by a learning process, and also use a pre-trained fixed network to reduce computational complexity. Experimental results demonstrate that the proposed method outperforms extant approaches in PSNR, computational complexity, and visual quality.
Reconstruction of missing color-channel data using a three-step back propagation neural network / J. Wang, M. Anisetti, G. Jeon. - In: INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS. - ISSN 1868-8071. - (2018). [Epub ahead of print]
Reconstruction of missing color-channel data using a three-step back propagation neural network
M. Anisetti;
2018
Abstract
Demosaicking aims to approximate missing color pixels through analysis of the geometric structure between given color pixels and missing color pixels. In this paper, we introduce an efficient adaptive demosaicking method based on back propagation (BP) neural network (BP-NN). We firstly reconstruct the green channel using one BPNN, and then refine the green channel utilizing another BPNN based on the color difference. With the whole green channel interpolated, we reconstruct the red/blue channel using the color difference between the green channel and red/blue channel in a local region. Finally, we refine the red/blue channel using the third BPNN. Regarding the interpolation issue, different image features have completely different properties, such as smooth regions, edges, and textures. Consequently, it is necessary to identify an adaptive model to estimate the relation among neighboring color pixels. We provide the adaptive BP-NN based demosaicking algorithm which can reduce blurring through recovery of missing pixels by a learning process, and also use a pre-trained fixed network to reduce computational complexity. Experimental results demonstrate that the proposed method outperforms extant approaches in PSNR, computational complexity, and visual quality.File | Dimensione | Formato | |
---|---|---|---|
Wang_et_al-2018-International_Journal_of_Machine_Learning_and_Cybernetics.pdf
accesso riservato
Tipologia:
Publisher's version/PDF
Dimensione
7.25 MB
Formato
Adobe PDF
|
7.25 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.