Sensing Matrix Design for Compressive Spectral Imaging via Binary Principal Component Analysis.

Area of Science:

• Optics and Photonics
• Image Processing
• Data Science

Background:

• Compressive spectral imaging (CSI) reduces spectral data-cube dimensions using coded-and-multiplexed projections.
• Traditional CSI sensing matrices can lose spectral data characteristics like variance.
• Principal component analysis (PCA) preserves data structure but its projection matrices are not binary for CSI.

Purpose of the Study:

• To develop an algorithm for designing binary sensing matrices for CSI that exploit PCA's structure-preserving properties.
• To improve spectral data reconstruction quality by maximizing data variance explanation in CSI acquisitions.

Main Methods:

• Estimating the spectral data covariance matrix from initial compressive measurements using random sensing matrices.
• Designing new binary sensing matrices by solving a non-convex optimization problem to approximate PCA principal components.
• Maximizing data variance explanation through the designed binary sensing matrices.

Main Results:

• Experimental validation demonstrates improved image reconstruction quality (up to 3 dB PSNR) using PCA-based binary sensing matrices.
• The proposed matrices outperform conventional random sensing matrices and existing PCA-based designed matrices.
• Enhanced preservation of spectral data structure and variance is achieved.

Conclusions:

• The developed algorithm effectively designs binary sensing matrices for CSI by integrating PCA principles.
• This approach significantly enhances spectral image reconstruction accuracy and fidelity.
• The method offers a practical solution for improving CSI system performance.