Segmentation of historical machine-printed documents using Adaptive Run Length Smoothing and skeleton segmentation paths

Image-based logical document structure recognition

“…Accepted covers define the page structure division. Run-Length Smearing Algorithm [30] is a well-known algorithm in this group based on segment growing.…”

“…The authors correct underand over-segmentation based on deducing the number of pictures. The work [29] presents a summary of segmentation methods and their ability to recognize various document structures.…”

“…For the Docstrum method [32], its default parameter values K ¼ 5 (the number of nearest neighbours), f t ¼ 2:578 (threshold of connecting components into line), f d ¼ 9 (proportion of letters), f pe ¼ 1:3 (threshold of connecting components into text blocks for perpendicular direction), f pa ¼ 1:5 (threshold of connecting components into text blocks for parallel direction) were substituted with K ¼ 8, f t ¼ 2:578, f pe ¼ 1:1, f pa ¼ 2:112. For ARLSA algorithm [30] the values of parameters were as follows T h ¼ 3:5 (threshold referring to the proportion of component height), c ¼ 0:4 (constant), a ¼ 5 (constant). T max (threshold referring to the number of congruent background pixels) was set as an average of all components.…”

“…T max (threshold referring to the number of congruent background pixels) was set as an average of all components. The remaining parameters were unchanged and were set as described in the paper [29]). The algorithm RXYC [21] was run with a window size of 16 Â 32.…”

Segmentation of Arabic Handwritten Documents into Text Lines using Watershed Transform

“…The first agent estimates global parameters and extracts the lines, the second one searches and detects adjacent components and the third one segments the touching lines. A similar technique is called Smearing method where the goal is to apply some transformation on the text bloc in order to group together homogeneous blocks constituting the lines [3,4,5,6].…”