Image Reconstruction from Incomplete Frequency Information Using Yang Method

Abstract

In many applications, an incomplete measurement case aims to obtain the desired original signal. However, the limited measured signal causes the predicted signal not to be the same as the original signal. A reconstruction technique is needed to improve the predicted signal. In this paper, we apply the Yang method for signal reconstruction from incomplete measurement, i.e. image reconstruction from incomplete frequency information. Low-high resolution patches of training images are learned to produce an overcomplete dictionary. Then the overcomplete dictionary is used to predict high-resolution patches on the testing images. Each testing images are targeted to reach the smallest RMSE. To obtain the smallest RMSE, each testing images have different conditions of variables, coming from the iteration number, the number of training images, and patch factor value. 0.3512 is the greatest RMSE improvement when comparing the smallest RMSE to the RMSE of the dirty image.