Figure 1

The blind compressed sensing (BCS) model decomposition: Here, few spatial weights and its corresponding temporal basis functions are shown. Note the weights have few non-zeros coefficients, and the learned temporal bases represent the temporal variations present in the data (eg: the second, and fourth example bases demonstrate the breathing motion, while the third demonstrate the faster cardiac pulsations.). The usage of such learned bases provide superior reconstructions compared to predetermined bases (such as Fourier bases) as demonstrated in Figure 2.