MRBENet: A Multiresolution Boundary Enhancement Network for Salient Object Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Salient Object Detection (SOD) aims to identify visually prominent objects in images, mimicking human perception.
• Convolutional Neural Network (CNN) based methods are effective but struggle with precise object boundaries and intact detection.
• Existing SOD models often fail to balance accurate localization with fine boundary detail preservation.

Purpose of the Study:

• To develop a novel network, the Multiresolution Boundary Enhancement Network (MRBENet), for improved salient object detection.
• To enhance the accuracy of salient object localization and the fineness of object boundaries.
• To address the limitations of current methods in preserving object details during detection.

Main Methods:

• Incorporated a deeper convolutional layer in the backbone for high-level semantic feature extraction and object localization.
• Employed a U-shaped network to extract edge features at multiple resolutions.
• Designed a Feature Fusion Module (FFM) to integrate edge and salient features, and a Feature Aggregation Module (FAM) with spatial attention for multiscale feature enhancement.

Main Results:

• The MRBENet successfully leverages edge features to optimize salient object location and boundary detail.
• Fusion of edge and semantic features via FFM and FAM significantly enhances salient feature representation.
• Experiments on six benchmark datasets confirm the method's high effectiveness and improved accuracy over state-of-the-art approaches.

Conclusions:

• The proposed MRBENet effectively improves salient object detection by enhancing boundary fineness and localization accuracy.
• The integration of multiresolution edge features and semantic information is crucial for robust SOD.
• MRBENet represents a significant advancement in salient object detection, outperforming existing methods on standard datasets.