Reconstruction method for Fourier single-pixel imaging by considering the distribution feature of Fourier spectra of images

The sparse sampling method based on Fourier single-pixel imaging technology has attracted a lot of attention, but there is a lack of a general reconstruction method for sparse sampling. To address this problem, a new reconstruction network is proposed with sparse sampling in Fourier domain. This method can achieve high-quality imaging with a low sampling rate and it can also suppress the ringing effect by predicting the frequency and spatial domain information. To preserve the image details, we consider the distribution of the Fourier spectra of images and design the two-dimensional mask and the loss function in the Fourier domain. Experiments show that when the image resolution is 256×256 and the sampling rate is as low as 8%, the reconstructed image can recover fine details of the object, and meanwhile suppress the ringing effect. More importantly, the reconstruction time of the proposed method is as low as 34ms, which is 4 orders of magnitude faster than the compressed sensing algorithm. This work can also be applied to other sparse sampling methods, which significantly promotes the process of practical application of single-pixel technology.