Text Extraction for Handwritten Circuit Diagram Images.

Paper-based handwritten electrical circuit diagrams still exist in educational scenarios and historical contexts. In order to check them or to derive their functional principles, they can be digitized for further analysis and simulation. This digitization effectively performs an electrical graph extraction and can be achieved by straight-forward instance segmentation, in which electrical symbols become the nodes and the interconnecting lines become the edges of the graph. For an accurate simulation however, the texts for describing the graph’s items properties have to be extracted as well and associated accordingly. The paper at hand describes a dataset as well as approaches based on optical character recognition, regular expressions and geometric matching. The source code of the described approach as well as the ground truth is integrated into a publicly available dataset and processing software.