Salient object detection based on multi-scale contrast.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Deep learning networks have significantly advanced salient object detection.
• Very deep convolutional networks are commonly used for feature extraction.
• Increasing network depth can lead to more training errors in deep learning models.

Purpose of the Study:

• To develop an improved salient object detection model.
• To address the challenge of increased training errors with deeper networks.
• To enhance detection accuracy by reducing image complexity and refining features.

Main Methods:

• Utilized residual networks to increase network depth while mitigating errors.
• Employed image simplification with color and texture features on multiple scales using super-pixels.
• Refined pixel-level features using a multi-scale feature correction method.
• Integrated multi-scale and multi-level features in a final fully connected layer for classification.

Main Results:

• The proposed model demonstrated superior performance compared to existing deep learning-based salient object detection methods.
• The integration of residual networks and image simplification effectively reduced complexity and improved accuracy.
• Pixel-level feature refinement successfully avoided errors introduced by region-level simplification.

Conclusions:

• The developed model offers a more accurate and robust approach to salient object detection.
• Combining residual networks with image simplification is a promising strategy for enhancing deep learning models.
• The method effectively handles feature complexity and improves the identification of salient targets.