Dynamic speckle imaging reconstruction via optical-flow-guided network

Scattering imaging aims to recover object images from complex scattered light fields and holds significant value in fields such as biological tissue imaging and non-line-of-sight (NLOS) detection. Although numerous methods have been developed for static object scattering imaging, moving objects induce additional temporal variations in speckle patterns, presenting significant challenges for high-quality imaging under dynamic conditions. To address this issue, this paper proposes a dynamic scattering imaging reconstruction method based on optical flow constraints. This method utilizes an optical flow network to extract motion information between consecutive speckle frames, treating it as an equivalent representation of the object's motion. Consequently, relying on the reconstruction of a single speckle frame, temporal prediction is achieved through optical flow warping. The optical flow network learns the motion field from speckle sequences in an unsupervised manner, eliminating the need for ground truth optical flow labels. Compared to frame-by-frame reconstruction methods, the proposed approach significantly enhances temporal consistency while maintaining high reconstruction accuracy. Experimental results demonstrate that the proposed method achieves high-quality dynamic reconstruction in practical motion scenarios, providing a novel approach for efficient imaging in complex scattering environments.