Internal micro-CT image enhancement method for rice seedling stems based on a pre-enhancement network and diffusion...

Area of Science:

• Biomedical Imaging
• Computer Vision
• Image Processing

Background:

• Micro-computed tomography (micro-CT) of rice seedling stems embedded in paraffin suffers from low contrast and lost detail.
• Existing image enhancement methods may not adequately address these specific challenges.

Purpose of the Study:

• To develop an advanced image enhancement technique for micro-CT scans of rice seedling stems.
• To improve image clarity, detail preservation, and hierarchical representation.

Main Methods:

• A novel method combining a pre-enhancement network with a denoising diffusion probabilistic model (DDPM).
• The pre-enhancement network utilizes multi-scale residual blocks (MSRB) with triple wavelet attention (TWA) and selective kernel feature fusion (SKFF).
• A variational information bottleneck (VIB) block is integrated into the DDPM to enhance noise prediction.

Main Results:

• The proposed method significantly enhances image contrast and detail compared to existing techniques.
• Experimental results show superior performance across objective assessment metrics.
• The enhanced DDPM model, conditioned on pre-enhancement features, produces CT images with improved hierarchical representation.

Conclusions:

• The integrated diffusion model and pre-enhancement network effectively addresses low contrast and lost detail in rice seedling stem micro-CT images.
• The approach offers a substantial improvement in computational efficiency (7.65x faster than conventional DDPM).
• This method provides strong technical support for advanced computer vision applications in biological imaging.