Estimation of disparity maps by compressive sensing

Compressive sensing enables the reconstruction of a signal from its small number of samples in a sparse domain. It is advantageous to use compressive sensing to achieve dense signals in situations where measurements are costly, as in the case of disparity maps. In this study, disparity values are reconstructed from samples taken of the ground truth values in frequency domain via Gaussian, Uniform distributions and along star-shaped 22 radial lines using total variation minimization. The results are compared in terms of accuracy and speed. The results of each method are shown with four commonly used images in the Middlebury dataset. The accuracies for the methods are changing according to the frequency content of the image used. The sampling matrix of 22 radial lines is the most successful among the methods proposed in this study in terms of speed and accuracy.