Feature enhanced non-line-of-sight imaging using graph model in latent space

Non-line-of-sight (NLoS) imaging reveals hidden scene from indirect diffusion signals. However, it is still challenging to balance noise suppression, detail preservation, and reconstruction efficiency. In this work, a robust framework which is centered on feature extractor and enhancement is proposed. In the framework, the feature extractor exploits the graph model in latent space for efficient noise suppression and detail preservation, the enhancement collaboratively learns the feature and data statistics by considering the extractor to define regularization. The reconstruction results on the publicly accessible datasets show that the proposed framework outperforms the state-of-art methods considering both quality and efficiency.