Deep Neural Network Based Salient Object Detection with Image Enhancement

Salient object detection aims to discover the most visually attractive regions from images. It allows more efficient follow-up processing of images without handling redundant information. In this paper, we propose a novel framework based on deep neural network to detect salient objects. The proposed framework introduces feature enhancement to input images to improve the performance of the fully convolutional neural network (FCN). Images are segmented and weighted through superpixel based pulse coupled neural networks. Low-level features including contrast and spatial features are extracted during this procedure by removing background disturbance in images. Subsequent neural network takes the enhanced images in and produces the saliency maps. Finally, some refinements are made afterwards to achieve better saliency results. Experimental results on five representative benchmarks show the superiority of our model than other state-of-the-art methods. Furthermore, comparisons are made to verify the effectiveness of image enhancement part in our model.