Gabor feature enhancement learning for speckle imaging through scattering media

Scattering medium imaging technology allows us to identify targets even in complex scattering environments, making it a widely researched area. In recent years, deep learning methods have shown impressive performance in this field. However, most existing approaches depend on training with pairs of target speckle data, and due to incomplete extraction of speckle features, their image restoration performance is often limited. To tackle these challenges, this article introduces a method known as the Gabor feature enhancement scattering imaging network (GFENet), which is designed to be feature-sensitive. Our approach begins by leveraging the autocorrelation characteristics of the scattering field as prior information in the reconstruction network. This process can enhance the network’s generalization capabilities and improves image reconstruction performance. Additionally, we implement a learnable Gabor filter that analyzes image features across multiple scales and directions, significantly boosting the network’s ability to extract speckle features. We have validated the effectiveness of our proposed method using multiple datasets.