MFC-Net : Multi-feature fusion cross neural network for salient object detection

Although methods based on the fully convolutional neural networks (FCNs) have shown strong advantages in field of salient object detection, existing still two challenging issues: insufficient multi-level feature fusion ability and boundary blur. To overcome these issues, we propose a novel detection method multi-feature cross network (denoted MFC-Net). Firstly, to issue ability, inspired by connection mode human brain neurons, framework, combined with contextual transfer modules (CFTMs) integrate, enhance transmit information an iterative manner. Secondly, address blurred boundaries, effectively edge features saliency map simple enhancement strategy. Thirdly, reduce loss caused generated fusion, use (FFMs) learn from multiple angles then output resulting map. Finally, hybrid function supervises at pixel level, optimizing performance. The proposed MFC-Net has been evaluated using five benchmark datasets. performance evaluation demonstrates that outperforms other state-of-the-art methods, which proves superiority approach.