Improved DCT-based nonlocal means filter for MR images denoising

Area of Science:

• Medical Imaging
• Image Processing
• Signal Processing

Background:

• Nonlocal Means (NLM) filters are effective for feature-preserved denoising in Magnetic Resonance (MR) images.
• Existing NLM filters can be further optimized for noise suppression efficiency.

Purpose of the Study:

• To develop a novel NLM filter utilizing Discrete Cosine Transform (DCT) for improved MR image denoising.
• To enhance the accuracy of similarity weight estimation in NLM filtering.

Main Methods:

• A new NLM filter is proposed, incorporating DCT for calculating similarity weights in the neighborhood's DCT subspace.
• The DCT's properties of low data correlation and high energy compaction are leveraged.
• Performance is evaluated against original NLM and Unbiased NLM (UNLM) filters.

Main Results:

• The proposed DCT-based NLM filter demonstrates superior denoising performance in MR images.
• Qualitative and quantitative evaluations confirm enhanced noise reduction capabilities.
• The method achieves more accurate weight estimation compared to traditional NLM approaches.

Conclusions:

• The DCT-based NLM filter offers a significant advancement in MR image denoising.
• This approach effectively suppresses noise while preserving crucial image features.
• The proposed method outperforms existing NLM techniques for MR image noise reduction.