Compressive sensing image recovery using dictionary learning and shape-adaptive DCT thresholding.

Area of Science:

• Medical Imaging
• Signal Processing
• Computer Vision

Background:

• Compressed sensing (CS) is effective for image recovery, relying on sparse priors.
• Dictionary learning (DL) offers adaptive bases but can blur details in complex image regions.
• Existing methods struggle with preserving fine structures in medical images.

Purpose of the Study:

• To propose a novel joint dictionary learning and Shape-Adaptive DCT (SADCT) method for CS medical image recovery.
• To enhance sparsity in shape-adaptive regions, particularly for complex medical image textures.
• To improve the reconstruction of image details and structures.

Main Methods:

• Jointly applying dictionary learning with SADCT thresholding.
• Exploiting local image sparsity within shape-adaptive regions.
• Utilizing attenuated SADCT coefficients for signal estimation.
• Developing an alternating minimization algorithm based on split Bregman iterations.

Main Results:

• The proposed method enhances sparsity, especially in detailed image areas.
• SADCT domain representation leads to sparser images with better detail preservation.
• Experimental results on MR images demonstrate efficient image recovery.
• The algorithm shows advantages over current leading CS reconstruction approaches.

Conclusions:

• The joint DL and SADCT method effectively recovers medical images using compressed sensing.
• This approach significantly improves the preservation of image details and structures.
• The proposed algorithm offers superior performance compared to existing CS reconstruction techniques.