Lowering latency and processing burden in computational imaging through dimensionality reduction of the sensing...

Area of Science:

• Computational imaging
• Signal processing
• Electromagnetics

Background:

• Frequency-diverse computational imaging systems simplify conventional architectures by moving constraints to the digital layer.
• Image reconstruction in these systems can be latency- and computationally intensive.

Purpose of the Study:

• To reduce latency and processing burden in frequency-diverse computational imaging.
• To propose a generic, unsupervised approach for optimizing image reconstruction.

Main Methods:

• Truncating insignificant principal components of the sensing matrix.
• Applying the approach directly to the sensing matrix without requiring training data or specific scene constraints.

Main Results:

• Reduced processing time and memory consumption in computational microwave and millimeter wave imaging.
• Minor impact on the quality of reconstructed images.
• Demonstrated feasibility in security screening applications.

Conclusions:

• The proposed method offers a new degree of freedom in image reconstruction, balancing performance between image quality and computational efficiency.
• This approach is essential for the widespread deployment of computational imagers in various scenarios.