Single Pixel Camera with Compressive Sensing by non-uniform sampling

Compressive Sensing refers to the acquisition in compressed form of sparse signals. One of the CS applications is the Single Pixel Camera that rethinks the conventional imaging systems. The key part of the camera is a Spatial Light Modulator (SLM), which provides the mechanism for taking arbitrary measurements of the scene. The non-uniform sampling (NUS) scenario offers a simple procedure to select random samples from the scene, instead of taking random projections. This paper analyses the Single Pixel Camera performance using NUS in the case of two categories of samples: randomly distributed over the scene and selected from the contours vicinity. The NUS is compared, in terms of reconstructed images, with measurements obtained by projection on a random selection from the S-Transform basis. The results are given on simulated data and on real images obtained with the experimental model of a Single Pixel Camera. The comparison of reconstructed images in the three cases of measurements shows that, in terms of PSNR, the NUS with contour pixels is equivalent with the S-transform based measurement.