| Title |
Single Image based Reflection Removal using Edge Separation |
| Authors |
박지민(Jimin Park) ; 이덕우(Deokwoo Lee) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.4.57 |
| Keywords |
Computer vision; Image processing; Deep learning; Unsupervised learning; Reflection removal |
| Abstract |
In this paper, a novel unsupervised learning-based method for effectively removing reflections from a single image is proposed. The method utilizes the edge intensity of a single image to classify the edges into strong and weak edges, applying a low-pass filter and a high-pass filter accordingly to generate a preprocessed image. The Double DIP model, a framework leveraging the intensity difference between two images, is employed to separate transparent layers, blur strong edges, and highlight weak edges to produce a single image. Subsequently, an improved transparent layer separation module, the Double DIP framework, separates the input and preprocessed images into transmissive and reflective layers. This iterative process gradually removes reflective components, enhancing the performance of reflection removal. Experimental results demonstrate that the proposed method outperforms existing unsupervised learning-based approaches for reflection removal and validates the feasibility of achieving effective reflection removal without labeled data. |