SRRV: A Novel Document Object Detector Based on Spatial-Related Relation and Vision

Abstract

Document object detection is a challenging task due to layout complexity and object diversity. Most of existing methods mainly focus on vision information, neglecting representative inherent spatial-related relationship between document objects. To capture structural information and contextual dependencies among document objects, we propose a novel framework combining spatial-related relation and vision (SRRV) for document object detection. SRRV consists of three parts: vision feature extraction network, relation feature aggregation network and result refinement network. Vision feature extraction network enhances information propagation of hierarchical feature pyramid by adopting feature augmentation paths. Then, relation feature aggregation network combines graph construction module and graph learning module. Specifically, graph construction module calculates spatial awareness information from geometric attributes to encode relationship, while graph learning module stacks Graph Convolutional Network (GCN) layers to aggregate relation information propagated among objects at global scale. Both the vision and relation features are fed into result refinement network for feature fusion and relational reasoning. Experiments on the PubLayNet, POD and Article Regions demonstrate that spatial relation information significantly improve the performance with better accuracy and precise bounding box prediction.