<論文>Complex-log Mappingを適用した特異値分解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 an important problem. In this paper, the recognition of rotated images is accomplished by applying singular value decomposition. That is to say, this method is made applying singular value decomposition to complex-log mapped images, which are generated as pre-processing from original images. Rotated images are transformed as parallel shift on the complex-log mapped plane, and also scaled images are transformed as the same phenomena in case of some original images. Extraction of features from complex-log mapped images is realized by applying the shift-invariant method. Generally, in case that the number of the pixels of the mapped image is M×N pieces with M≧N, the number of singular values in N pieces at most. The statistical data, which are calculated from singular values, show propriety of the proposed method. Finally recognition ratios are calculated by experiments with a neural network. This neural network is 3-layer hierarchical one with a hidden layer.
東海大学の論文
2002-09-30