The Multi-Focus-Image-Fusion Method Based on Convolutional Neural Network and Sparse Representation

Area of Science:

• Image Processing
• Computer Vision
• Artificial Intelligence

Background:

• Multi-focus image fusion is essential in image processing.
• Sparse Representation (SR)-based and Convolutional Neural Network (CNN)-based methods are widely used but have limitations.
• SR methods are local with nonlinear rules, while CNN methods are global with linear rules.

Purpose of the Study:

• To propose a novel fusion method combining CNN and SR for improved image information.
• To leverage the strengths of both SR and CNN for more precise and abundant fused image details.
• To reduce computational complexity compared to existing state-of-the-art methods.

Main Methods:

• A new fusion method is proposed, integrating CNN to assist SR.
• Source image patches are fused using SR guided by weights generated by a CNN.
• The method aims for a global fusion approach with enhanced local feature preservation.

Main Results:

• The proposed method significantly outperforms existing SR and CNN-based fusion methods.
• Experimental results show superior visual perception and objective evaluation metrics.
• The method achieves a substantial reduction in computational complexity.

Conclusions:

• The novel CNN-assisted SR method offers superior performance in multi-focus image fusion.
• This approach provides a more effective and computationally efficient solution for image fusion tasks.
• The findings suggest a promising direction for future research in image processing and computer vision.