A combined HMM-PCNN model in the contourlet domain for image data compression

Area of Science:

• Image processing and computer vision
• Signal processing
• Machine learning for data compression

Background:

• Multiscale geometric analysis (MGA) offers advantages over wavelet transforms for representing complex image features like edges and contours.
• Existing MGA-based image compression research is limited due to data structure differences and coding complexity.
• JPEG2000, a wavelet-based standard, has limitations in representing high-dimensional singular data.

Purpose of the Study:

• To propose a novel MGA-based image compression scheme.
• To address the challenges in MGA-based image coding.
• To improve compression performance and encoding flexibility.

Main Methods:

• Utilizing contourlet transform for sparse image decomposition, capturing multiscale and multidirectional characteristics.
• Employing the Hidden Markov Model (HMM) to model relationships between contourlet coefficients.
• Integrating the Pulse-Coupled Neural Network (PCNN) for state probability classification within the HMM framework.
• Using the Expectation-Maximization (EM) algorithm for HMM training.

Main Results:

• The proposed HMM/PCNN-contourlet model demonstrates superior compression performance compared to existing methods.
• The model achieves better representation of image features like edges and contours.
• The encoding scheme offers enhanced flexibility.

Conclusions:

• The HMM/PCNN-contourlet model is an effective approach for image data compression.
• This method overcomes limitations of traditional wavelet-based compression for complex image data.
• Further research into MGA-based coding schemes is warranted.