Recently, many salient object detection methods have utilized edge contours to constrain the solution space. This approach aims to reduce the omission of salient features and minimize the inclusion of non-salient features. To further leverage the potential of edge-related information, this paper proposes a Redundant and Missing Feature Decoupling Network (RMFDNet). RMFDNet primarily consists of a segment decoder, a complement decoder, a removal decoder, and a recurrent repair encoder. The complement and removal decoders are designed to directly predict the missing and redundant features within the segmentation features. These predicted features are then processed by the recurrent repair encoder to refine the segmentation features. Experimental results on multiple Red–Green–Blue (RGB) and Red–Green–Blue-Depth (RGB-D) benchmark datasets, as well as polyp segmentation datasets, demonstrate that RMFDNet significantly outperforms previous state-of-the-art methods across various evaluation metrics. The efficiency, robustness, and generalization capability of RMFDNet are thoroughly analyzed through a carefully designed ablation study. The code will be made available upon paper acceptance.