Deep Saliency with Encoded Low Level Distance Map and High Level Features

By Gayoung Lee, Yu-Wing Tai, Junmo Kim. Computer Vision and Pattern Recognition, 2016

464 citations recorded by the source corpus.

Recent advances in saliency detection have utilized deep learning to obtain high level features to detect salient regions in a scene. These advances have demonstrated superior results over previous works that utilize hand-crafted low level features for saliency detection. In this paper, we demonstrate that hand-crafted features can provide complementary information to enhance performance of saliency detection that utilizes only high level features. Our method utilizes both high level and low level features for saliency detection under a unified deep learning framework. The high level features are extracted using the VGG-net, and the low level features are compared with other parts of an image to form a low level distance map. The low level distance map is then encoded using a convolutional neural network(CNN) with multiple 1 1 convolutional and ReLU layers. We concatenate the encoded low level distance map and the high level features, and connect them to a fully connected neural network classifier to evaluate the saliency of a query region. Our experiments show that our method can further improve the performance of state-of-the-art deep learning-based saliency detection methods.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/b1ec1f1515eb29b9f2b26280b8daffe0d8804215

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Reverse Attention for Salient Object Detection (2018, methodology): “Maximum F-Measure of recent deep CNN-based saliency detection models on ECSSD, including DS [6], ELD [7], DCL [8], DHS [8], RFCN [9], NLDF [10], DSS [11], MSRNet [12], Amulet [13], UCF [14], and ours (red circle).”
A Simple Pooling-Based Design for Real-Time Salient Object Detection (2019, background): “[6] combined low-level heuristic features, such as color histogram and Gabor responses, with high-level features extracted from CNNs.”
Deep Level Sets for Salient Object Detection (2017, methodology): “Qualitative comparison with FT [1], HC [10], GMR [56], BSCA [47], MTDS [31], MDF [29], MCDL [62], ELD [26], LEGS [50].”
A Stagewise Refinement Model for Detecting Salient Objects in Images (2017, methodology): “Ours BL DCL DHS DRFI DS ELD KSR LEGS MCDL MDF RFCN Time (s) 0.07 31.73 0.41 0.04 46.21 0.12 0.57 50.90 1.54 2.27 21.55 4.72 (i) (ii) (iii) (iv) 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Precision Recall Fmeasure (i) (ii) (iii) (iv) 0.4 0.5 0.6 0.7 0.8 0.9 Precision Recall Fmeasure Figure 9.”
Object Detection With Deep Learning: A Review (2018, background): “[149], MTDNN [141], CRPSD [142], DCL [143], ELD [153], NLDF [154] and DSSC [155].”
Pyramid Feature Attention Network for Saliency Detection (2019, methodology): “…0.8114 0.0654 0.9046 0.0622 0.8901 0.0532 0.8456 0.0960 - - DSS[15] 0.8135 0.0646 0.8959 0.0647 0.9011 0.0476 0.8506 0.0998 0.7603 0.0751 ELD[18] 0.7372 0.0924 0.8674 0.0811 0.8409 0.0734 0.7882 0.1228 0.7195 0.0909 NLDF[24] 0.8125 0.0648 0.9032 0.0654 0.9015 0.0481 0.8518 0.1004 0.7532…”
Detect Globally, Refine Locally: A Novel Approach to Saliency Detection (2018, methodology): “[16] utilize a two-stream framework with high-level feature descriptors extracted from the VGG-net and low-level heuristic features such as color histogram and Gabor responses.”
Salient Object Detection in the Deep Learning Era: An In-Depth Survey (2019, background): “• ELD [60] concatenates deep convolution features and an encoded low level distance map (ELD-map) to construct a feature vector for each super-pixel.”
Attentive Feedback Network for Boundary-Aware Salient Object Detection (2019, result): “We compared our algorithm with other 12 state-ofthe-art deep learning methods, which are LEGS [26], RFCN [28], ELD [13], DCL [16], DS [18], DHS [21], Amulet [33], DSS [10], C2SNet [17], RAS [5], DGRL [29] and PiCANet [22].”
Contrast Prior and Fluid Pyramid Integration for RGBD Salient Object Detection (2019, background): “[34] considered both high-level feature extracted from CNNs and hand-crafted feature.”
A Bi-Directional Message Passing Model for Salient Object Detection (2018, methodology): “Secondly, early works [29, 18, 15, 12] predict saliency maps by mainly using high-level features from deep convolutional layers.”
Salient Object Detection With Pyramid Attention and Salient Edges (2019, methodology): “For example, some methods integrate deep learning models with hand-crafted features [20], heuristic saliency priors [36], level set [15], contextual information [57], or explicit visual ﬁxation [40].”
Salient object detection: A survey (2014, background): “[63] exploit two subnetworks to encode lowlevel and high-level features separately.”
Structure-Measure: A New Way to Evaluate Foreground Maps (2017, background): “, 2016),DHS (Liu&Han, 2016), and ELD (Lee et al., 2016)] to generate the binary and non-binary foreground maps.”