Reconstruction of simulated VLBI data using the SARA method and random raw patches

VLBI (very long baseline interferometry) is used to image astronomical objects. However, the image quality decreases due to incomplete visibility measurements. The visibility is affected by the frequency, baseline length, measurement duration, and some telescopes used. A popular method that can solve this quality degradation problem in VLBI is SARA (sparsity averaging reweighted analysis). This study aims to increase the SARA performance with scoped in improving NRMSE (normalized root mean square error) and SNR (signal-to-noise ratio) on simulated VLBI data. We used random raw patch pairs from the SARA results to get a trained dictionary. In collecting the data for training, we used ten original images of the surrounding objects that we captured. Then we measured it using a VLBI sampling pattern, processed it using SARA, and applied the SARA results as a pre-reconstruction image. After that, we trained these ten pairs of the original and pre-reconstruction images. In reconstruction, we used the dictionary to test two simulated data and compare the results with the SARA. Finally, the results showed that the combination of SARA and random raw patches had a smaller NRMSE and a higher SNR than the SARA alone.