An Effective Headlines Extraction Method Employing a basis Function Network
T. Sato, Tsutomu Miki · Intelligent Automation & Soft Computing · 2006
Abstract This paper describes an effective feature extraction method by using a basis function network. The basis function network, which we have proposed [1][2], decomposes an original image into weighted basis functions corresponding to spatial frequency. Local image corresponding to a specific spatial frequency can be picked up by reconstructing with selected basis functions. We apply the method to extract headline areas in a document. Evaluation function for assigning the weights is specialized for the headline azeas extraction. The proposed method has an advantage that the extraction results are independent of a document orientation. Extractions of headlines from Japanese and English newspapers with various orientations are examined and the validity of the proposed method is discussed.