A fast and effective model for wavelet subband histograms and its application in texture image retrieval

Area of Science:

• Digital image processing
• Signal processing
• Computer vision

Background:

• Wavelet transforms are widely used for image compression and analysis.
• Existing image retrieval methods often require uncompressed data or have high computational costs.
• Compression domain analysis offers potential for efficient retrieval.

Purpose of the Study:

• To propose a novel and efficient method for characterizing wavelet subbands.
• To develop a new image signature based on bit-plane probabilities.
• To enable effective image retrieval in the compression domain, particularly for JPEG2000 images.

Main Methods:

• Characterization of wavelet subbands through bit-plane extraction.
• Association of each bit plane with the probability of 1-bit occurrence.
• Concatenation of bit-plane probabilities to form a unique image signature.
• Direct signature extraction from the code-block code-stream.

Main Results:

• The proposed image signatures require less storage and computational power.
• Experimental results demonstrate superior cost-effectiveness compared to existing methods.
• The method shows high efficacy in texture image retrieval.
• Signatures are adaptable for compression domain image retrieval (e.g., JPEG2000).

Conclusions:

• The bit-plane probability-based signature is an effective and efficient method for image characterization.
• This approach offers significant advantages for image retrieval in compressed formats.
• The proposed method is a cost-effective alternative to current state-of-the-art techniques.