Overlaid Arrow Detection for Labeling Regions of Interest in Biomedical Images

Abstract

A new template-free geometric signature-based technique detects arrow annotations on biomedical images. Arrow detection is a key first step to region-of-interest (ROI) labeling and image content analysis. Images are first binarized using a fuzzy binarization tool, and candidates are selected based on the connected component principle. For each candidate, the proposed method checks geometric properties of novel arrow signatures from key points associated with its boundary. These signatures are then compared with the theoretical (or idealized) arrow signatures, and a high similarity score indicates the presence of an arrow. The data was evaluated against the imageCLEFmed benchmark collection and achieved precision and recall of 93.14 percent and 86.12 percent, respectively, which outperforms previously reported arrow-detection methods.