Compressive Spectral Light Field Image Reconstruction via Online Tensor Representation.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Computer Vision

Background:

• Spectral light field (SLF) imaging captures 5D data (2D spatial, 2D angular, 1D spectral).
• Existing SLF sensors are complex, costly, and require time-consuming, independent spectral and light field capture.
• Spatio-spectral and angular information is crucial for advanced applications like microscopy and computer vision.

Purpose of the Study:

• To propose a compressive spectral light field imaging architecture for efficient data acquisition.
• To develop a computational algorithm for reconstructing 5D SLF information from compressed measurements.
• To offer an efficient alternative to conventional SLF imaging methods.

Main Methods:

• A compressive imaging architecture is proposed to capture multiplexed SLF data with fewer measurements.
• A tensor-based computational algorithm is developed to recover 5D information by exploiting high correlations in SLF data.
• Tucker decomposition is applied to compressed measurements for online calculation of an ad-hoc dictionary basis, enhancing reconstruction accuracy.

Main Results:

• Simulations and experimental results demonstrate the proposed method's efficiency compared to conventional techniques.
• The compressive imaging device and algorithm significantly reduce the number of required measurements for SLF capture.
• The tensor-based algorithm shows lower computational complexity than matrix-based methods, enabling faster processing.

Conclusions:

• The proposed compressive spectral light field imaging architecture and tensor-based reconstruction algorithm offer an efficient and fast alternative for capturing and processing multidimensional images.
• This approach overcomes limitations of conventional SLF sensors by integrating spectral and light field capture.
• The method enables accurate reconstruction and faster processing, advancing applications in microscopy, computer vision, and beyond.