Industrial Scene Text Detection With Refined Feature-Attentive Network

Detecting the marking characters of industrial metal parts remains challenging due to low visual contrast, uneven illumination, corroded character structures, and cluttered background part images. Affected by these factors, bounding boxes generated most existing methods locate low-contrast text areas inaccurately. In this paper, we propose a refined feature-attentive network (RFN) solve inaccurate localization problem. Specifically, design parallel feature integration mechanism construct an adaptive representation from multi-resolution features, which enhances perception multi-scale texts at each scale-specific level generate high-quality attention map. Then, attentive refinement is developed map rectify location deviation candidate boxes. addition, re-scoring designed select with best rectified location. Moreover, two scene datasets, including total 102156 images 1948809 instances various structures parts. Extensive experiments on our dataset four public datasets demonstrate that proposed method achieves state-of-the-art performance.