An image understanding system using attributed symbolic representation and inexact graph-matching.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Traditional image analysis often struggles with global information comprehension.
• Representing complex image features and their relationships is a significant challenge.

Purpose of the Study:

• To develop a powerful image understanding system using a semantic-syntactic representation.
• To enable robust analysis and interpretation of global image content.

Main Methods:

• Utilized attributed relational graphs (ARGs) where nodes represent global image features and branches represent relations.
• Employed a multilayer graph transducer scheme for hierarchical symbolic mapping from spatial to global representation.
• Implemented dynamic programming for calculating ARG distances and inexact matching to handle noise and distortion.

Main Results:

• Successfully extracted ARG representations from images.
• Defined a distance measure between images based on their ARG representations.
• Demonstrated system capabilities in object localization and target detection in SAR images.

Conclusions:

• The proposed system offers a robust approach to image understanding via attributed relational graphs.
• The system effectively handles real-world image challenges like noise and distortion.
• The methodology shows promise for advanced image analysis tasks.