tailieunhanh - Báo cáo hóa học: " Nonlinear Image Restoration Using a Radial Basis Function Network"

Tuyển tập báo cáo các nghiên cứu khoa học quốc tế ngành hóa học dành cho các bạn yêu hóa học tham khảo đề tài: Nonlinear Image Restoration Using a Radial Basis Function Network | EURASIP Journal on Applied Signal Processing 2004 16 2441-2450 2004 Hindawi Publishing Corporation Nonlinear Image Restoration Using a Radial Basis Function Network Keiji Icho The Research Development Department Matsushita Electric Industrial Co. Ltd. Osaka 571-8501 Japan Email Youji Iiguni Department of Systems Innovation Graduate School of Engineering Science Osaka University Osaka 560-8531 Japan Email iiguni@ Hajime Maeda Department of Communications Engineering Graduale School of Engineering Osaka University Osaka 565-0871 Japan Email maeda@ Received 4 August 2003 Revised 22 April 2004 We propose a nonlinear image restoration method that uses the generalized radial basis function network GRBFN and a regularization method. The GRBFN is used to estimate the nonlinear blurring function. The regularization method is used to recover the original image from the nonlinearly degraded image. We alternately use the two estimation methods to restore the original image from the degraded image. Since the GRBFN approximates the nonlinear blurring function itself the existence of the inverse of the blurring process does not need to be assured. A method of adjusting the regularization parameter according to image characteristics is also presented for improving restoration performance. Keywords and phrases radial basis function network regularization nonlinear image restoration steepest descent technique alternate estimation. 1. INTRODUCTION In the recent years a special class of artificial neural networks called the radial basis function network RBFN has received considerable attention 1 . The RBFN has a universal approximation capability 2 and has successfully been applied to many signal and image processing problems due to its excellent approximation capability 3 4 . The RBFN provides a smooth function that achieves a good tradeoff between fidelity to the data and smoothness. The regularization parameter .