Bayesian recognition of local 3-d shape by approximating image intensity functions with quadric polynomials.

Area of Science:

• Computer Vision
• Image Processing
• Geometric Modeling

Background:

• Complex object recognition and pose estimation are crucial in computer vision.
• Previous methods often require significant computational resources for initial shape analysis.

Purpose of the Study:

• To develop a computationally efficient method for recognizing basic 3-D surface shapes (spheres, cylinders, planes) from image patches.
• To establish a foundational step for subsequent complex object recognition and parameter estimation.

Main Methods:

• Image partitioning into small windows, each representing a surface patch.
• Approximation of image data within windows using 2-D quadric polynomials.
• Classification based on hypotheses of planar, cylindrical, or spherical surfaces.
• Low-pass filtering integrated into the approximation process.

Main Results:

• A computationally simple shape recognition technique was developed.
• The method effectively classifies image patches as parts of spheres, cylinders, or planes.
• The approach offers approximate Bayesian minimum-probability-of-error recognition for large windows.

Conclusions:

• The proposed shape recognition method provides a computationally efficient and effective first step for advanced image analysis.
• Accurate classification of basic shapes enables subsequent detailed estimation of object parameters like shape, orientation, and location.