CVC-UAB's Participation in the Flowchart Recognition Task of CLEF-IP 2012

The aim of this document is to describe the methods we used in the flowchart recognition task of the CLEF-IP 2012 track. The flowchart recognition task consisted in interpreting flowchart linedrawing images. The participants are asked to extract as much as structural information in these images as possible and return it in a predefined textual format for further processing for the purpose of patent search. The Document Analysis Group from the Computer Vision Center (CVC-UAB) has been actively working on Graphics Recognition for over a decade. Our main aim in participating in the CLEF-IP flowchart recognition task is to test our graphics recognition architectures on this type of graphics understanding problem. Our recognition system comprises a modular architecture where modules tackle different steps of the flowchart understanding problem. A text/graphic separation technique is applied to separate the textual elements from the graphical ones. An OCR engine is applied on the text layer while on the graphical layer identify with nodes and edges as well as their relationships. We have proposed two different families of node and edge segmentation modules. One dealing with the raw pixel data and another working in the vectorial domain. The locations of nodes identified are fed to the recognizer module which is in charge of categorizing the node’s type. We have proposed two different node descriptors for the recognizer module. The module analyzing the edges is analysing the connections between nodes and categorizes the edge style. Finally, a post-processing module is applied in order to correct some syntactic errors. We have submitted four different runs by combining the two variants of the segmentation module together with the two variants of the recognition module.