Salient Object Detection via Integrity Learning

Motivation

To facilitate integrity learning for salient object detection (SOD), we design a novel Integrity Cognition Network (ICON), which explores three important components to learn strong integrity features.

Key Components

1. Diverse Feature Aggregation (DFA): Unlike the existing models that focus more on feature discriminability, we introduce a diverse feature aggregation component to aggregate features with various receptive fields (i.e., kernel shape and context) and increase the feature diversity. Such diversity is the foundation for mining the integral salient objects.

2. Integrity Channel Enhancement (ICE): Based on the DFA features, we introduce the integrity channel enhancement component with the goal of enhancing feature channels that highlight the integral salient objects (i.e., micro and macro levels) while suppressing the other distracting ones.

3. Part-Whole Verification (PWV): After extracting the enhanced features, the part-whole verification method is employed to determine whether the part and whole object features have strong agreement. Such part-whole agreements can further improve the micro-level integrity for each salient object.

Advantages

The proposed ICON framework addresses the critical challenge of detecting salient objects with complete structures. By explicitly modeling integrity at both the feature diversity level and the part-whole relationship level, ICON achieves more robust and accurate salient object detection, especially for objects with complex structures or in cluttered backgrounds.

The three components work synergistically: DFA provides diverse features as the foundation, ICE selects the most relevant channels for integrity, and PWV ensures consistency between parts and the whole, leading to high-quality detection results with strong object integrity.