Document Line Segmentation Based on Wavelet Transform

In this paper a novel approach to document lines segmentation is presented. The algorithm is based on wavelet transform of the horizontal projective profile of the document image. The projective profile is examined as a one-dimensional signal which is decomposed using the pyramidal wavelet algorithm up to a precise level, where local minima and maxima are discovered. These local extrema, projected into the input signal, correspond to the blank spaces between document lines and to the pivots of the lines. The method is tested to a broad set of printed and handwritten documents and has proved to be stable and efficient.