In order to optimize the utility of anti-reflection removal in images,this paper proposes a novel deep learning network for anti-reflection removal based on visual attention mechanism soft pooling.The network architecture is simple and efficient,and by integrating visual attention soft pooling theory and flash image information without interference,it a-chieves accurate estimation of the transmission image,thereby efficiently eliminating reflections.In a wide range of real-world scene tests,this method not only outperforms several mainstream anti-reflection techniques in terms of visual effects,but also demonstrates outstanding performance in objective evaluation indicators of image quality after anti-reflec-tion removal.
维普
万方数据