36
Figure 4.3 : As the number of measurements (M) used in CS increases, the mean-
squared error (MSE) between the magnitudes of the reconstructed image and the
reference image for the object in Figure 4.2, normalized by the squared energy of the
reference image, decreases. The MSE decay is fast from 50 to IOO CS measurements.
After the number of CS measurements exceeds the “sparsity” of the object, the decay
flattens out.
Due to the reduced rate of convergence of the min-TV reconstruction algorithm for
number of CS measurements below 250, the reconstruction uses the minimization of
the Zl-norm in the wavelet basis [8] for the results in Figure 4.3. The normalized
MSE of each reconstructed image is computed as follows,
MSE =
(4.1)
where the summations are across all image pixels, Xr is the reconstructed image and
X is a 32 × 32 greyscale reference image downsampled from a digital photograph of
the object. The MSE sharply decreases from 50 to 100 CS measurements and then
decays more slowly afterwards. The shape of this decay curve is dependent on the