Patch-primitive driven compressive ghost imaging.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Statistical Learning

Background:

• Ghost imaging (GI) has advanced significantly but faces limitations in practical applications due to low reconstruction quality and high measurement requirements.
• Natural images possess inherent structural simplicity and shared primitives within local image patches.
• Existing GI reconstruction methods often struggle to leverage these intrinsic image properties effectively.

Purpose of the Study:

• To enhance the reconstruction quality of ghost imaging by addressing limitations in measurement requirements and image fidelity.
• To develop a novel ghost imaging reconstruction approach driven by learned image patch primitives.
• To integrate local image structure priors into the compressive ghost imaging framework.

Main Methods:

• A patch-primitive driven reconstruction approach utilizing statistical learning.
• Sparse representation of image patches using an over-complete dictionary of learned primitives.
• Incorporation of local image priors into a convex optimization framework for compressive ghost imaging.

Main Results:

• The proposed method achieves superior reconstruction quality compared to existing methods using the same number of measurements.
• The approach effectively reduces the number of requisite measurements for obtaining satisfactory ghost imaging results.
• Experimental validation demonstrates the efficacy of the patch-primitive driven strategy.

Conclusions:

• The patch-primitive driven reconstruction significantly improves ghost imaging performance.
• This method offers a pathway to overcome practical limitations of ghost imaging by optimizing measurement efficiency and reconstruction fidelity.
• The integration of learned image priors represents a promising direction for advanced computational imaging techniques.