Angular pattern and binary angular pattern for shape retrieval.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Shape retrieval is crucial for content-based image retrieval systems.
• Existing shape descriptors often face challenges with scale, rotation, and computational efficiency.
• Global descriptors offer efficiency but sometimes lack discriminative power compared to local descriptors.

Purpose of the Study:

• To propose two novel, intrinsically scale and rotation invariant shape descriptors: angular pattern (AP) and binary angular pattern (BAP).
• To develop a multiscale integration of AP and BAP for enhanced shape retrieval performance.
• To evaluate the computational efficiency and discriminability of the proposed descriptors against state-of-the-art methods.

Main Methods:

• Development of Angular Pattern (AP) and Binary Angular Pattern (BAP) as global shape descriptors.
• Implementation of a multiscale integration strategy for combining AP and BAP features.
• Experimental validation using the MPEG-7 and TARI-1000 benchmark datasets.

Main Results:

• AP and BAP demonstrate intrinsic invariance to scale and rotation.
• The proposed global descriptors achieve high discriminability, comparable to state-of-the-art local descriptors.
• Experiments show superior effectiveness compared to existing shape retrieval methods on benchmark datasets.

Conclusions:

• The proposed AP and BAP descriptors provide an efficient and effective solution for shape retrieval.
• Multiscale integration of AP and BAP enhances performance for real-world applications.
• The method presents a computationally attractive alternative for large-scale shape retrieval tasks.