Abstract

In shadow detection area, deep learning-based methods have achieved outstanding performance. However, the potential ambiguity in shadow images is still the main reason for the poor performance of the algorithms. By analysing the relationships among shadow edges, shadow false-positive information, and shadow areas, an edge-guided disambiguation module for better shadow detection performance is proposed. The edge-guided disambiguation module consists two parts: 1) the shadow edge feature is used to guide the comprehensive shadow area feature generation of different resolution. 2) The shadow false-positive features is applied to reduce ambiguity in the comprehensive shadow area features. In this way, the missed detections is added in the first step and the false detections is reduced in the second step. More accurate detection results is obtained. The comprehensive experiments are proceeded on the three public shadow detection datasets: SBU, UCF, and ISTD. The experimental results demonstrated the effectiveness and robustness of the proposed method.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call