A Thinning-based Junction Detection And Resolution Algorithm for Document Images
Hasan S. M. Al-Khaffaf, Abdullah Zawawi Talib, Universiti Sains Malaysia · The Journal of The University of Duhok · 2020
Junction detection plays a significant role in document image recognition.High recognition rate of graphical primitives is correlated with the proper detection of junctions.In this paper, a junction detection algorithm is presented in thinning-based raster to vector conversion process.The method has three stages that leverage junction representation from pixels to features (i.e.junctions).The input image is thinned to its skeleton.Edges were found next and pixels with many neighbours are designated as a low level junction.Polygonal approximation on edges is used to detect L-junctions while connected component analysis is used to find intermediate junctions.Intermediate junctions with a distance less than a threshold are combined to form a high level X-and Y-junctions.Performance evaluation on mechanical engineering drawings shows precision rate of 82.38% and a recall rate of 97.29%.