Using an exact performance of Hough transform for image text segmentation

The (/spl rho/,/spl theta/)-interpretation of Hough transform (HT) is treated herein considering its facilities to localize straight lines or cut-offs in a given image. The definition of "exact" HT is introduced for the image and for the given choice of the respective HT-domain size (/spl rho//sub size/, */spl theta//sub size/). The idea of "HT inner noise" is also defined using a mean square estimate for the HT-deviations from the exact HT. Hence, a constructive method is proposed for a set of algorithms realising digital HT of an arbitrary small inner noise. The experiment of algorithms and a comparative analysis of results are also represented. The proposed method and algorithms are applied for pre-processing of hand-written text images. Besides the localisation of text rows and words, the approach proposed is especially effective for evaluation of the averaged letter slope of handwriting in the case of critical resolution of 2/spl divide/3 pixels per line and/or per distance among lines.