Touching Component Extraction and Separation in Arabic Manuscripts

This paper addresses the problem of touching characters (TC) extraction and separation in Arabic manuscripts. It proposes a recognition based method to separate them and join each component piece to its corresponding word. The proposed method extracts the TCs in the document whether between successive text-lines or words of the same text-line. For vertically/up-down TCs, we improved an existing method. For horizontally /left-right TCs, we proposed a novel extraction method based on the morphology analysis of the terminal letters of Arabic words. Then, it recognizes the TCs relying on templates, using shape context descriptor and an interpolation function the TPS transformation (Thin Plate Spline). Finally, it segments them based on the distance from the central points (midpoints or gravity centers) of the recognized template’s parts. Tests are performed using a large dataset of TCs and three metrics: Manhattan, Euclidean and Canberra distances. Obtained results strongly support the efficiency of the proposed TC extraction and segmentation methods and outperform results of some related works, taking into account the different types, variability and complexity of the TCs.