A Conventional Framework for Reflection Removal Using Single Images.

Reflections caused by poor lighting conditions or the presence of transparent media such as glass often degrade image quality by introducing unwanted visual artifacts and obscuring important scene information. These reflections not only reduce the perceptual quality of images but also adversely affect the performance of downstream computer vision applications. Reflection removal is therefore a critical pre-processing step for enhancing image quality and improving the reliability of vision-based systems. In this paper, we propose a pipeline framework that employs individual components from established prior works on reflection removal. The proposed method uses a hierarchical thresholding consisting of two stages to extract a reflection-free structure layer. DCT and soft matting suppress reflection components in the texture layer while preserving important image details. The performance of the proposed approach was evaluated on a benchmark dataset developed by ROSE Lab at Nanyang Technological University, Singapore. Experimental results, assessed using multiple quantitative and perceptual quality metrics, demonstrate that the proposed method effectively suppresses reflections, produces images with fewer unnatural artifacts, and achieves improved perceptual quality relative to the baseline methods evaluated in the study, based on no-reference metrics' results. The competitive visual quality and natural appearance of the restored images suggest the proposed framework may be a promising candidate for pre-processing in computer vision applications.