A Self-organizing Feature Map for Arabic Word Extraction

Arabic word spotting is a key step for Arabic NLP and the text recognition task. Many recent studies have addressed segmentation problems in the Arabic language. However, many issues still have to be overcome. In this paper, we propose a new approach for segmenting an image Arabic text into its constituent words. Our approach consists of two main steps. In the first step, a set of features is extracted from connected components using the Run-length smoothing algorithm (RLSA). In the second step, spatially close connected components that are likely to belong to the same word component are grouped together. This is done via a learning technique called the self-organizing feature map (Kohonen map). We evaluated our approach on 300 images with different sizes and fonts for handwritten text using AHDB. Our results suggest that our approach can efficiently segments lines. Moreover, as our approach is based on a straightforward machine learning model, it should be possible to adapt it to other languages as well.