Unsupervised Multimodal Remote Sensing Image Registration via Domain Adaptation

Shi, Lukui; Zhao, Ruiyun; Pan, Bin; Zou, Zhengxia; Shi, Zhenwei

doi:10.1109/tgrs.2023.3333889

Cited by 1 publication

(1 citation statement)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Jin et al [1]introduced generative adversarial networks specifically crafted to counteract the adverse effects of atmospheric turbulence. In a noteworthy contribution, Li et al [34] provided a comprehensive definition of the non-rigid distortions induced by turbulence, characterizing them as deformable grids. Subsequently, they employed a grid deformer for estimating the distortion field and an image generator for crafting distortion-free images.…”

Section: Registration-fusion Approachmentioning

confidence: 99%

Deformation-aware image restoration from atmospheric turbulence based on quasiconformal geometry and pulse-coupled neural network

Yin,

Sun,

Zhang

et al. 2024

Preprint

View full text Add to dashboard Cite

Atmospheric turbulence is a major factor in image degradation issues such as blurring, distortion and intensity fluctuations when monitoring long-range targets. The randomness, spatiotemporal variation and perturbations of turbulence make it challenging to restore vision-friendly and credible images from degraded image sequences. In this work, we address the problem by proposing a deformation-aware image restoration algorithm based on quasiconformal geometry and pulse-coupled neural network (PCNN). To accurately measure the magnitude of geometric deformation caused by turbulence, the deformation within degraded images is specified in a non-conformal distortion that disrupts local geometry. The Beltrami coefficient uniquely associated with the quasiconformal maps is applied to quantify the average distortion degree. The deformation-aware measurement minimizes registration errors in aligning degraded images by more reliable reconstruction of reference frames. Additionally, an improved PCNN model inspired by the primary visual cortex is developed to boost the perceptual quality of the restored image with lucky image fusion. The absence of manual parameter tuning and the ability to simultaneously process image sequences in the PCNN model enhance the robustness of the restoration algorithm. The performance of our algorithm is validated by experiments on physically simulated and real data, which contain 220 sequences with 22928 frames. The results show that our algorithm can yield a superior restoration through atmospheric turbulence compared with several state-of-the-art methods. The code is available at https://github.com/whuluojia/ImTurb.

show abstract

Section: Registration-fusion Approachmentioning

confidence: 99%