A visual-attention model using Earth Mover's Distance-based saliency measurement and nonlinear feature combination.

Area of Science:

• Computer Vision
• Computational Neuroscience
• Image Processing

Background:

• Traditional visual attention models often rely on Difference-of-Gaussian filters.
• Combining features in existing models can be suboptimal.
• There is a need for advanced models for both static and dynamic saliency mapping.

Purpose of the Study:

• To introduce a novel computational visual-attention model.
• To improve the accuracy of static and dynamic saliency maps.
• To offer a biologically inspired approach to visual attention.

Main Methods:

• Utilized Earth Mover's Distance (EMD) for center-surround difference calculation, replacing Difference-of-Gaussian filters.
• Implemented a two-step nonlinear feature combination process: Lm-norm for super-feature creation and Winner-Take-All for final combination.
• Extended the model to generate dynamic saliency maps using spatiotemporal receptive fields and EMD.

Main Results:

• The proposed model achieved superior performance on both static image and video datasets.
• EMD proved effective in measuring center-surround differences in visual attention.
• The biologically inspired feature combination enhanced model capabilities.

Conclusions:

• The novel visual-attention model offers a significant advancement over existing methods.
• The use of EMD and specific nonlinear operations provides a more effective approach to saliency mapping.
• The model's success in both static and dynamic scenarios highlights its versatility and robustness.