Optical twin of the Laplacian pyramid for multi-scale feature extraction in efficient image perception
Year: 2026
Type: article
Abstract: With the development of advanced imaging technology, multiple models and strategies have been introduced to process images of different dimensions and scales, encompassing microscopic images, hyperspectral images, and remote sensing images. For large-scale image perception, such as remote sensing imagery, image pyramids serve as an efficient digital processing technique to extract multi-scale features. However, the high resolution and multi-layer computations bring extra burdens for computational hardware. In the information era, optical computing emerges as a novel computational paradigm for parallel and efficient computations in the visual perception chain. Leveraging the advantages of optics and light waves, we propose an optical twin of the Laplacian pyramid (OLP) for multi-scale feature extraction in large-scale image processing and visual intelligence. The OLP scheme is established based on the deduction of the equivalence between Laplacian pyramid layer computations and frequency domain filtering. We successively establish the optical implementation of the digital Laplacian pyramid and conduct experiments to validate the effectiveness of the OLP scheme. The OLP method achieves a classification accuracy of 98.75% and 94.87% on the UC Merced and AID dataset, respectively, while transferring 80.24% of the computational burden from electronics to optics.
Source: Optics Express
DOI: 10.1364/oe.610274
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,270
Award: G3715892857
OpenAlex ID: W7204248449
