Affine-similar shape retrieval: application to multiview 3-D object recognition

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Representing 3-D objects using a limited set of standard views is crucial for efficient object identification.
• Curvature Scale Space (CSS) image contours are a standardized method for 2-D shape representation, effective under similarity transforms.

Purpose of the Study:

• To investigate the feasibility of representing 3-D objects using a few standard 2-D views.
• To evaluate the effectiveness of Curvature Scale Space (CSS) contours for 3-D object recognition under affine transformations.

Main Methods:

• Object boundaries are treated as 2-D shapes and represented by maxima locations in their CSS image contours.
• The study analyzes the impact of affine transformations on CSS maxima locations.
• A shape-based retrieval system is applied to multiview 3-D object representation and recognition using silhouette contours of 18 aircraft models.

Main Results:

• The locations of CSS image maxima exhibit stability even under significant affine transformations.
• The proposed system demonstrated robust and efficient 3-D free-form object recognition using multiview representation.
• The CSS representation proved effective for identifying objects from random views using stored standard view descriptions.

Conclusions:

• Multiview 3-D object representation and recognition can be reliably achieved using the Curvature Scale Space (CSS) method.
• The CSS representation is a viable technique for handling affine transformations in 3-D object recognition tasks.
• This approach offers a robust solution for shape-based retrieval in large image databases.