Abstract Objectives: This study focuses on developing a hybrid detection model that can improve document element like signature detection, by combining singlestage and two-stage object detection architectures. The objective is to improve the detection of objects with higher processing efficiency. Method: YOLOv8n is used for fast and real-time detection, and Faster R-CNN is used for accurate classification. Post-processing method: The approach involves the combination of predictions with their respective confidence scores and IoU values. Findings: The overall model is more accurate with average precision of 100.00%, recall of 92.15%, F1-score of 95.74%, IOU of 76.29%, and Dice co-efficient of 83.77% than the individual models, especially for challenging document categories. Novelty: This work introduces a new custom dataset and the presentation of a hybrid detection model that combines YOLOv8n and Faster R-CNN for improved document element detection. In this proposed system, instead of relying on either a speed-optimized or accuracy-optimized model, the proposed system combines fast initial detection with region-based refinement in a single model. The proposed system is particularly beneficial for semistructured document organization and can be applied to real-world signature detection problems. Keywords: Document Element Detection; YOLOv8n; Faster R-CNN; Hybrid Model; Signature Detection; Object Detection
Thejashwini et al. (Thu,) studied this question.