アフィン項を導入した動径基底関数ネットワークによる画像の補間

概要

論文の詳細を見る
Interpolation of digital images is one of the important problems in the fields of image processing, for example, image enlargement and geometric correction of remote sensing. Because image information as two-dimensional signals is possibly nonlinear, then linear interpolation of images cannot efficiently solve the interpolation problems. Although there are several interpolation method, such as nearest neighbor interpolation (NNI), bi-linear interpolation (BLI) and cubic Convolution interpolation (CCI), one of the popular methods in the field of nonlinear functional approximation is back propagation learning in neural networks. However, back propagation learning in neural networks requires larger amount of computational time than radial basis function (RBF) networks. In this paper, we apply standard radial basis interpolation adopted affine terms (ARBI) to digital image data interpolation and enlargement problems, because it is natural to consider both linear and nonlinear of image information as two-dimensional signals. We apply thin-plate spline functions to ARBI. Then, we evaluate interpolation performance between radial basis interpolation and other methods by calculation signal-to-noise ratios. As a result, it is shown that interpolation with radial basis function networks adopted affine terms exhibits good signal-to-noise ratios (SNRs) as well as other methods.
社団法人電子情報通信学会の論文
1997-08-13