Document segmentation using polynomial spline wavelets

Wavelet transforms have been widely used as effective tools in texture segm entation in the past decade. Segmentation of document images, which usually contain three types of texture information: text, picture and background, can be regarded as a special case of texture segmentation. B-spline wavelet s possess some desirable properties such as being well localized in time an d frequency, and being compactly supported, which make them an effective to ol for texture analysis. Based on the observation that text textures provid e fast-changed and relatively regular distributed edges in the wavelet tran sform domain, an efficient document segmentation algorithm is designed via cubic B-spline wavelets. Three-means or two-means classification is applied for classifying pixels with similar characteristics after feature estimati on at the outputs of high frequency bands of spline wavelet transforms. We examine and evaluate the contributions of different factors to the segmenta tion results from the viewpoints of decomposition levels, frequency bands a nd wavelet functions. Further performance analysis reveals the advantages o f the proposed method. (C) 2001 Pattern Recognition Society. Published by E lsevier Science Ltd. Ail rights reserved.