A generic shape/texture descriptor over multiscale edge field: 2-D walking ant histogram

Area of Science:

• Computer Science
• Image Processing
• Multimedia Systems

Background:

• Multimedia databases require efficient content-based retrieval methods.
• Existing shape descriptors often struggle with complex object boundaries and textures.
• Automated feature extraction is crucial for large-scale multimedia indexing.

Purpose of the Study:

• To present a novel, automated shape descriptor for multimedia content-based retrieval.
• To demonstrate the descriptor's capability as an efficient texture descriptor.
• To evaluate the descriptor's performance within the MUVIS framework for multimedia indexing and retrieval.

Main Methods:

• A multiscale approach is applied to edge fields to extract significant object boundary subsegments.
• Object boundaries are traversed, and local features are described using 2-D histograms.
• The descriptor is integrated as a feature extraction module into the MUVIS framework.

Main Results:

• The proposed method successfully extracts relevant edge segments representing major object boundaries.
• The descriptor demonstrates superior performance for directional textures, functioning effectively as a texture descriptor.
• Integration into the MUVIS framework facilitated testing of its performance in multimedia indexing and retrieval.

Conclusions:

• The novel shape descriptor offers an effective solution for content-based retrieval in multimedia databases.
• Its dual capability as a texture descriptor enhances its versatility and performance.
• The method provides a robust feature extraction module for advanced multimedia indexing and retrieval systems.