In computer vision, the reflections impair the overall perception of images and introduce undesired interference to high-level image understanding. To date, reflection removal approaches have been developed with advantage of visual quality enhancement. However, the reflection removal for single image is still a challenging task with its performance to be further boosted. During this process, additional visual artifacts such as blocking may also be introduced that degrades subjective quality of images. In this paper, we make the first attempt to eliminate image reflections and artifacts simultaneously within a unified framework. A structure-texture strategy is utilized to decompose the image into structure and texture layers. The proposed model handles reflection removal in both layers and further suppress blocking artifacts in the texture layer. Finally, a fusion process is employed to obtain the result image. Experiments with objective and subjective evaluations validate the superiority of the proposed method, which outperforms several state-of-the-art algorithms.

在计算机视觉中，反射会损害图像的整体感知力，并对高级图像理解造成不希望的干扰。迄今为止，已经开发了具有视觉质量增强优势的反射消除方法。但是，要进一步提高其性能，单张图像的反射去除仍然是一项艰巨的任务。在此过程中，可能还会引入其他视觉伪像（例如遮挡），从而降低图像的主观质量。在本文中，我们首次尝试在统一框架内同时消除图像反射和伪影。利用结构纹理策略将图像分解为结构层和纹理层。所提出的模型处理两层中的反射消除，并进一步抑制纹理层中的块状伪影。最后，采用融合过程获得结果图像。通过客观和主观评估的实验证明了该方法的优越性，该方法优于几种最新算法。