US20250317658
2025-10-09
Electricity
H04N23/95
The innovation described involves a diffractive camera capable of class-specific imaging, which uses deep learning-structured transmissive surfaces to capture images of target object classes while erasing other classes. This system operates all-optically, meaning it performs these tasks without requiring additional digital processing. The camera's design leverages optical mode filtering achieved through 3D-assembled substrate layers, optimized to form clear images of target objects and erase non-target objects from the output view.
This technology falls within the optical deep learning domain, where complex functions are executed at the speed of light. The camera employs an optical neural network to selectively image target object classes and erase non-target classes. This approach addresses privacy concerns by preventing raw image data from being captured and subsequently processed, reducing the risk of exposure through adversarial attacks or data breaches.
Traditional digital cameras and computer vision techniques have raised privacy concerns due to continuous image data collection. Existing solutions often rely on post-processing algorithms or hardware-level interventions to modify images after capture, which can be computationally intensive and environmentally taxing. These methods also risk exposing original data during transmission or storage. The diffractive camera aims to address these issues by enforcing privacy before digitization, reducing computational demands and enhancing data security.
The diffractive camera uses a series of optimized diffractive layers to filter optical modes, ensuring only target class images are formed at the output plane while others are erased. This passive system requires no external computing power beyond illumination light, making it highly efficient. Experimental demonstrations with THz radiation and 3D-printed layers showed successful class-specific imaging, even with complex datasets like the MNIST handwritten digits.
This technology can be extended across different parts of the electromagnetic spectrum for various applications, such as privacy-preserving digital cameras and task-specific imaging systems. By providing a power-efficient solution that minimizes data storage and transmission needs, it aligns with global demands for sustainable and secure imaging technologies. Its ability to perform class-specific encryption further enhances its utility in sensitive environments.