Speckle-Driving De-Artifact Nets ghost imaging

Different than the conventional imaging methods, ghost imaging (GI) tests the response of an object to different spatial spectrums, which brings potential advantages. However, the application of GI still faces some thorny issues. One of the major challenging is building a light source of high-modulation-rate, high-resolution and low cost. Sparse array is a feasible solution, but the artifacts caused by side lobes highly deteriorate the image quality. We here propose an architecture for GI called Speckle-Driving De-Artifact Nets (SDDAN), which maps the relationship between artifacts and speckles. The experimental demonstrations show that SDDAN can effectively remove the artifacts in reconstructed images and the peak signal to noise ratio is 10.80 dB and 8.75 dB higher than compressed sensing and denoising convolutional neural networks. SDDAN makes a step to applications of GI and may be a general method for scenes involving sparse array.

Automated summary

Different from conventional imaging methods, ghost imaging (GI) tests the response of an object to different spatial spectrums, which has potential advantages. However, there are still challenges in the application of GI, such as the need for a high-modulation-rate, high-resolution, and low-cost light source. This study proposes a architecture called Speckle-Driving De-Artifact Nets (SDDAN) for GI, which effectively removes artifacts in reconstructed images and achieves higher peak signal-to-noise ratio compared to other methods.