Infrared image super-resolution method for edge computing based on adaptive nonlocal means

2021 
Infrared images generally suffer from low resolutions, which restricts the further data mining of these images. Super-resolution (SR) technology effectively improves the spatial resolution of the infrared image without changing the existing hardware imaging equipment. Therefore, it is a promising computational imaging approach with resource-limited edge devices. In this paper, an adaptive-threshold nonlocal means (NLM)-based SR algorithm is proposed. Specifically, an image quality assessment index of infrared SR results is designed and introduced into the NLM reconstruction algorithm. On the one hand, it is used as a threshold to determine the iterative convergence condition of the algorithm; on the other hand, it is used as an evaluation standard to select the best reconstructed HR image among multiple output results. A GPU acceleration strategy is also proposed to ensure the high efficiency of the edge computing process for reducing the computational time. Experimental results demonstrate that the algorithm realizes the adaptive iteration of the NLM-based SR of the infrared images, and it significantly improves the geometric structures and detail recognition of the original input LR images. The sawtooth and ringing effects of the algorithm are less, and its objective evaluation indexes are also significantly improved compared with those of other SR algorithms.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    40
    References
    0
    Citations
    NaN
    KQI
    []