Learning to Sense for Coded Diffraction Imaging.

Area of Science:

• Computational imaging
• Image reconstruction
• Signal processing

Background:

• Coded diffraction imaging (CDI) is a powerful technique for reconstructing images from diffraction intensity measurements.
• Phase retrieval remains a critical challenge in CDI, often requiring iterative algorithms with significant computational cost.
• Existing methods struggle to optimize illumination patterns, limiting reconstruction quality.

Purpose of the Study:

• To develop a framework for learning illumination patterns to enhance signal recovery in coded diffraction imaging.
• To improve image reconstruction quality while maintaining a fixed computational cost.
• To demonstrate the effectiveness of learned illumination patterns using a limited training dataset.

Main Methods:

• An alternating minimization-based phase retrieval method was employed as the iterative algorithm.
• The iterative phase retrieval process was modeled as an unrolled neural network with a fixed number of layers.
• Illumination patterns were optimized to minimize the signal recovery error.

Main Results:

• The proposed method significantly improves image reconstruction quality compared to traditional approaches.
• Learned illumination patterns enhance signal recovery at a fixed computational cost.
• Effective learning of illumination patterns was achieved using only a small number of training images.

Conclusions:

• The framework effectively learns illumination patterns for superior coded diffraction imaging.
• This approach offers a computationally efficient solution for high-quality image reconstruction.
• The method shows promise for various applications requiring precise signal recovery from diffraction data.