Document image segmentation based on wavelet features

This paper proposes a method for segmentation of images containing both textual and graphical data. The method uses wavelet transformation to build the feature vector and a pattern recognition technique to classify areas of a document image. Values of wavelet coefficients distribution histogram of the source images sliding window serve as elements of a feature vector. For recognition of document area category, we use a trained classifier based upon support vector machine with RBF kernel.