Salient object detection is a fundamental problem and has been received agreat deal of attentions in computer vision. Recently deep learning modelbecame a powerful tool for image feature extraction. In this paper, we proposea multi-scale deep neural network (MSDNN) for salient object detection. Theproposed model first extracts global high-level features and contextinformation over the whole source image with recurrent convolutional neuralnetwork (RCNN). Then several stacked deconvolutional layers are adopted to getthe multi-scale feature representation and obtain a series of saliency maps.Finally, we investigate a fusion convolution module (FCM) to build a finalpixel level saliency map. The proposed model is extensively evaluated on foursalient object detection benchmark datasets. Results show that our deep modelsignificantly outperforms other 12 state-of-the-art approaches.
展开▼
