PuSH - Publikationsserver des Helmholtz Zentrums München

Khan, A.H.* ; Umer, R.M. ; Dunnhofer, M.* ; Micheloni, C.* ; Martinel, N.*

LBKENet: Lightweight Blur Kernel Estimation Network for Blind Image Super-Resolution.

In: (Image Analysis and Processing – ICIAP 2023: 22nd International Conference, ICIAP 2023, 11-15 September, Udine, Italy). Berlin [u.a.]: Springer, 2023. 209-222 (Lect. Notes Comput. Sc. ; 14234 LNCS)
DOI
Blind image super-resolution (Blind-SR) is the process of leveraging a low-resolution (LR) image, with unknown degradation, to generate its high-resolution (HR) version. Most of the existing blind SR techniques use a degradation estimator network to explicitly estimate the blur kernel to guide the SR network with the supervision of ground truth (GT) kernels. To solve this issue, it is necessary to design an implicit estimator network that can extract discriminative blur kernel representation without relying on the supervision of ground-truth blur kernels. We design a lightweight (LBKENet) approach for blind super-resolution (Blind-SR) that estimates the blur kernel and restores the HR image based on a deep convolutional neural network (CNN) and a deep super-resolution residual convolutional generative adversarial network. Since the blur kernel for blind image SR is unknown, following the image formation model of the blind super-resolution problem, we first introduce a neural network-based model to estimate the blur kernel. This is achieved by (i) a Super Resolver that, from a low-resolution input, generates the corresponding SR image; and (ii) an Estimator Network generating the blur kernel from the input datum. The output of both models is used in a novel loss formulation. The proposed network is end-to-end trainable. The methodology proposed is substantiated by both quantitative and qualitative experiments. Results on benchmarks demonstrate that our computationally efficient approach (12 $$\times $$ fewer parameters than the state-of-the-art models) performs favorably with respect to approaches that have less number of parameters and can be used on devices with limited computational capabilities.
Altmetric
Weitere Metriken?
[➜Einloggen]
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Konferenzbeitrag
Korrespondenzautor
Schlagwörter Anisotropic Blur Kernels ; Blind Image Super-resolution (blind-sr); ; Isotropic Blur Kernel
ISSN (print) / ISBN 0302-9743
e-ISSN 1611-3349
Konferenztitel Image Analysis and Processing – ICIAP 2023: 22nd International Conference, ICIAP 2023
Konferzenzdatum 11-15 September
Konferenzort Udine, Italy
Zeitschrift Lecture Notes in Computer Science
Quellenangaben Band: 14234 LNCS, Heft: , Seiten: 209-222 Artikelnummer: , Supplement: ,
Verlag Springer
Verlagsort Berlin [u.a.]
Nichtpatentliteratur Publikationen
Institut(e) Institute of Computational Biology (ICB)