Multi-queue merging scheme and its application in Arabic script segmentation

It is an important topic for segmentation on how to merge the over-segmented subimages of characters into integral characters. The conventional character segmentation has been mostly "one dimensional" but this work attacks two-dimensionally complex character layout. For that, a multi-queue model is proposed. First, a positional relationship graph (PRG) is produced among graphemes, based on the positional constraints. With the PRG, several parallel queues of graphemes are generated. The merging problem is inherently a process of walking in segmentation space, which is generated from the multi-queue structure. With some defined metric, the optimization can be conducted to get the appropriate merging path. We test our model in Arabic script, and experiment results show the potential of our model