A robust text line detection in complex handwritten documents

In this paper, we present the modified method of detecting text lines in handwritten documents based on the Block-Based Hough Transform. The algorithm has the practical application in the manuscript author identification. The proposed technique consists of three steps: preprocessing, detecting of potential text lines and eliminating the false ones. The first step covers the following operations: image binarization, extraction of connected components and selection of supporting connected components based on the local maximal values in the vertical histogram strips. In the second step we select the appropriate subset of connected components supplemented by one-point components. Next, we use the block-based Hough transform to detect potential text lines. Finally, we detect possible false alarms (lines detected incorrectly). The proposed method is applied to the text lines analysis in the fifteenth century Latin manuscripts. Our approach was verified to be more effective than the traditional ones, in the best cases by twenty percent.