Compressive phase object classification using single-pixel digital holography.

Area of Science:

• Computational Imaging
• Holography
• Machine Learning

Background:

• Single-pixel cameras (SPCs) offer compressed sensing for image classification, avoiding intensive reconstruction.
• Direct inference from compressed signals is efficient but typically limited to intensity images.

Purpose of the Study:

• To extend direct inference for phase object classification using single-pixel digital holography (SPDH).
• To improve classification accuracy through joint optimization of sampling patterns and classifiers.

Main Methods:

• Acquiring compressed measurements of complex amplitudes via SPDH.
• Training a classifier directly on these compressed SPDH measurements.
• Jointly optimizing SPDH sampling patterns and the classification model.

Main Results:

• Successfully classified phase object images of handwritten digits from the MNIST database.
• Demonstrated the capability of SPDH for direct phase object classification, a challenge for conventional SPCs.
• Achieved improved classification accuracy via joint optimization.

Conclusions:

• SPDH provides an effective method for direct phase object classification using compressed measurements.
• The proposed joint optimization strategy enhances the performance of computational imaging systems for phase retrieval and classification.
• This technique offers a computationally efficient alternative for analyzing phase information in various applications.