1 June 2025 in APL Photonics
Non-line-of-sight (NLOS) imaging aims to reconstruct objects beyond line-of-sight view, enabling potential applications in various fields. However, a fundamental problem is that acquiring the transient images is time-consuming, making it challenging to balance reconstruction quality and acquisition efficiency. In this paper, the under-scanning NLOS imaging model is simplified into a convolution approximation model, and a computationally efficient deconvolution optimization-based NLOS imaging (DO-NLOS) method is proposed to provide noise-robust reconstructions from under-scanning measurements. Specifically, we incorporate Poisson probability distribution characteristics and a total variation norm of transient images as well as the non-negativity and sparseness of objects into the deconvolution process. This incorporation effectively compensates for the inaccuracies in deconvolution caused by under-scanning measurements and also facilitates the use of regularization terms in the deconvolution process. Through comprehensive evaluation of synthetic and experimental data, we have demonstrated that DO-NLOS can achieve high-quality reconstruction under extreme conditions, such as spatially scanned data and data with short exposure times. In addition, it exhibits the ability to reconstruct complete objects under under-scanning conditions as low as 0.146%. Even with an extremely short exposure time of 1.28 s, DO-NLOS can still distinguish objects with a distance of 6 cm, which is close to the axial resolution limit of the system. Our framework has great potential in the application of real-time scanning NLOS imaging.
4 July 2025 in Optics and Lasers in Engineering
24 December 2025 in Optics and Lasers in Engineering
12 January 2026 in Optics and Lasers in Engineering
7 September 2026 in Applied Physics Letters
Existing methods are often limited by their underlying physical models, making it challenging to achieve single-shot, non-invasive multispectral imaging through scattering media. In this Letter, a speckle correlation model for the single-shot imaging of multispectral targets is developed. By using a gradient descent method based on differentiable computation, efficient multispectral imaging through scattering media is achieved. The method requires no additional system modulation and uses only minimal computational resources for fast reconstruction. Reconstruction results obtained by conventional methods and by the proposed method are compared under increasing spectral differences. Additionally, the proposed method can image hidden targets with spectral differences of up to 160 nm.