Efficient texture image retrieval using copulas in a Bayesian framework.

Area of Science:

• Image processing and computer vision
• Statistical modeling
• Bayesian inference

Background:

• Content-based image retrieval (CBIR) is crucial for organizing and searching large image datasets.
• Existing methods often struggle to capture complex relationships within image features.
• Wavelet transforms are effective for multi-resolution image analysis.

Purpose of the Study:

• To develop a novel joint statistical model for subband coefficient magnitudes from the dual-tree complex wavelet transform (DT-CWT).
• To integrate this model within a Bayesian framework for improved content-based image retrieval.
• To evaluate the model's effectiveness for color texture retrieval.

Main Methods:

• Developed a joint statistical model for DT-CWT subband coefficient magnitudes.
• Incorporated modeling of marginal coefficient distributions.
• Coupled the model with a Bayesian framework for image retrieval.
• Conducted experiments on four color texture image databases.

Main Results:

• The proposed joint model effectively captures associations between transform coefficients across scales and color channels.
• Achieved competitive retrieval accuracy compared to state-of-the-art methods.
• Identified a model configuration offering low storage requirements and efficient runtime.
• Demonstrated applicability for large-scale image database deployment.

Conclusions:

• The novel joint statistical model offers a robust approach for content-based image retrieval.
• The model balances retrieval accuracy, computational efficiency, and storage needs.
• This work advances the field of CBIR, particularly for color texture analysis.