Abstract
Salient-object detection is a fundamental and the most challenging problem in computer vision. This paper focuses on the detection of salient objects, especially in low-contrast images. To this end, a hybrid deep-learning architecture is proposed where features are extracted on both the local and global level. These features are then integrated to extract the exact boundary of the object of interest in an image. Experimentation was performed on five standard datasets, and results were compared with state-of-the-art approaches. Both qualitative and quantitative analyses showed the robustness of the proposed architecture.
Highlights
Human beings very and quickly make vision-based decisions by extracting all important information from any real scene
The input image is split into small regions, and a Convolutional neural networks (CNNs) is used to extract features that are passed to multilayer perceptron (MLP) to compute the saliency of the region
The deep saliency models included nonlocal features (NLDF) [29], contour to saliency (C2S) [11], visual saliency detection based on multiscale deep CNN features (MDF) [37], deep saliency with encoded low-level distance map and high-level features (ELD) [30], salient-object detection in low-contrast images via global convolution and boundary refinement (GCBR) [38], and aggregating multilevel convolutional features for salient-object detection (Amulet) [39]
Summary
Human beings very and quickly make vision-based decisions by extracting all important information from any real scene. Sci. 2020, 10, 8754 is determined by the global contrast of image instead of local features, it becomes hard for the model to examine detailed boundary knowledge of the object To overcome these problems, we propose a boundary-aware fully convolutional network for the detection of salient objects that captures both the local and global context with a built-in refinement module to achieve segmentation with fine boundaries. We propose a boundary-aware fully convolutional network for the detection of salient objects that captures both the local and global context with a built-in refinement module to achieve segmentation with fine boundaries Integration of both local and global features helps our model to accurately locate salient region. The rest of this paper is organised as follows: Section 2 details the literature review, Section 3 explains the architecture of the proposed approach, Section 4 presents the experimental results, and Section 5 draws the conclusions
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have