Complex-log Mappingによる回転画像の認識

概要

論文の詳細を見る
In the field of digital image recognition, the extraction of rotation-invariant features is a very important problem. In this paper, the recognition of rotated images is made by applying the method of complex-log mapping. The system which is the pre-processing for image recognition consists of two stages. At the first stage, complex-log mapped images are produced from original input images. At the second stage, Fourier transform is applied to these mapped images. Although the final images produced at this second stage have been dealt with as rotation-and-scale invariant images, the discussion in this paper points out these final images, which are generally effective enough for rotation invariant images, are not necessary effective for scale invariant images. This system mentioned above is tested using binary images and those deteriorated by noise. To classify input images, the final data are fed into a neural network using error-back-propagation learning.The effectiveness against rotated images is shown by several kinds of computer simulations.
東海大学の論文