On the Regularization of a Prior-Knowledge-Based Estimator

• Main Authors: Yi-Chen Su, 蘇怡真
• Other Authors: Hsin-Chun Yu
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/34085296992893854164

碩士 === 東海大學 === 資訊管理學系 === 99 === Prior discrete Fourier transform(PDFT) has been applied successfully to various problems in image reconstruction. The PDFT solves the best approximation of the object to be reconstructed in a suitably designed Hilbert-Space with minimum-weighted energy. This way can effectively overcome the non-uniqueness problem of reconstructing the object from its limited measurements in spectrum and also improve the resolution quality. The PDFT algorithm usually uses the rectangular prior function to characterize the location and the size of the object. However, in the matrix operation, this way can be easily affected by the ill-posed problem and increase the difficulty of inversed reconstruction process. Moreover, it cannot be recovered into a satisfied image result if it was interfered with the noise. In order to expand the applicability of PDFT in the real image problem, the study aims to investigate the image performance and the ability of noise-resistance of the PDFT with different prior functions. We also use Miller-Tikhonor regularization and Eigen-decomposition to further explain the simulation results of different prior functions.