Predicting Illusory Contours Without Extracting Special Image Features.

Area of Science:

• Computer Vision
• Computational Neuroscience
• Image Processing

Background:

• Object boundary detection is crucial for computer vision, but real-world images often present incomplete or unclear boundaries.
• Illusory contours pose a challenge, requiring models to infer boundaries without explicit gradients, unlike traditional methods relying on specific features like junctions.

Purpose of the Study:

• To develop a robust object boundary detection model capable of handling incomplete and illusory contours.
• To propose a functional optimization approach that accounts for fundamental object properties without requiring specialized feature extraction.

Main Methods:

• A functional optimization approach assigns a cost to boundary arrangements, minimizing it to find the most probable interpretation.
• The model encodes object properties like alignment, continuity, and simplicity within the cost functional.
• Boundary arrangements are represented by border ownership maps, and multiple interpretations are handled via functional local minima.

Main Results:

• The model successfully detects boundaries for both real and illusory contours.
• It accurately predicts multiple interpretations, including illusory and real contours in the Kanizsa illusion.
• The approach avoids complex feature extraction, relying on basic image edges.

Conclusions:

• The proposed functional optimization model offers a robust solution for object boundary completion, including challenging illusory contour scenarios.
• This proof-of-concept model demonstrates the efficacy of encoding basic object properties for boundary detection.
• Future work may extend the model to more complex, real-world image conditions beyond synthetic, grayscale images.