SALIENCY DETECTION VIA DENSE CONVOLUTION NETWORK

Zheng Fang, Tieyong Cao and Yibo Xing

ABSTRACT: Saliency detection is a fundamental problem in the field of image processing and computer vision. The convolutional model has been also used in saliency detection for its outstanding performance on image classification and localization task. In this paper, we propose a novel way to detect the salient object by modifying the Dense Convolution Network (DenseNet). We replace the fully-connected layer and the final pooling layer in DenseNet into a convolution layer and a deconvolution layer to fit the saliency detection task. And our network ends up with a squared Euclidean loss layer for saliency regression. Our network is end-to-end architecture which outputs saliency maps directly. Experimental results demonstrate that our approach is competitive in comparison with the state-of-the-art approaches.

Keywords: Saliency detection, Convolution Neural Network, Dense Convolution Network, computer vision, end-to-end