Determining and Modifying Saliency of a Visual Medium

This invention uses computer modeling, user feedback, and eye tracking to identify key parts of images by evaluating curvature, color, and lighting, then dynamically adjusting these elements to enhance the visual focus.

Background

Methods that rely on lighting, texture, and color tend to be view-dependent and can disrupt the natural flow of a scene. As 3D content creation expands in interactive environments like video games and virtual reality, there is a need for a solution that manipulates attention without forcing the viewer’s experience. By using geometric features, this technology pushes the boundaries of visual design, ensuring that attention is guided more subtly and effectively.

Technology

Researchers at the University of Maryland have created a method of modifying 3D geometry to persuade visual attention. By altering the saliency of specific regions on 3D models, this method changes how viewers perceive these objects. The process works through mesh filtering and adjustments to a model’s geometry, enhancing features that are critical for drawing attention. Unlike traditional methods reliant on color or lighting, this solution operates on a deeper level by adjusting the physical structure of the model itself. Eye-tracking studies show that these changes successfully shift and hold viewer focus in a non-intrusive manner.

Benefits

  • Enhanced accuracy in detecting and classifying visually important regions
  • Integration of computer modeling, user input, and eye-tracking for a robust analysis
  • Dynamic adjustment of visual features to optimize saliency across different regions
  • Improved image segmentation that supports better automated processing and user interactions

Applications

  • Advertising saliency targeting
  • UI design optimization
  • Video engagement analytics
  • Visual content analysis
  • Digital marketing enhancement

Additional info

This system employs computer modeling, user input, and eye-tracking to calculate visual saliency from multiple channels. It analyzes metrics such as mean curvature, its gradients, color intensity gradients, luminance, color opponency, saturation, lighting, and focus. Additionally, it computes vertex-specific curvature changes, adjusting region parameters based on aggregated measurements to accurately classify and modify regions within a visual medium.