"AI-Designed Camera Only Records Objects of Interest While Being Blind to Others"

Digital cameras have been widely adopted in various aspects of society and are now widely used in mobile phones, security surveillance, autonomous vehicles, and facial recognition. Huge amounts of image data are generated by these cameras, thus raising concerns about privacy protection. Some existing methods address these concerns by employing algorithms, such as image blurring or encryption, to conceal sensitive information from images. However, these methods still expose sensitive data because raw images are captured before being digitally processed to hide or encrypt sensitive information. Furthermore, the computation of these algorithms necessitates an increase in power consumption. Other efforts have also been made to find solutions to this problem, such as using customized cameras to reduce image quality so that identifiable information can be hidden. Yet, these approaches sacrifice overall image quality for all objects of interest, which is undesirable, and they are still vulnerable to adversarial attacks that retrieve the recorded sensitive information. A new study published in eLight demonstrated a new paradigm for achieving privacy-preserving imaging by developing a fundamentally new type of imager designed by Artificial Intelligence (AI). Researchers from the University of California, Los Angeles (UCLA), led by Professor Aydogan Ozcan, presented a smart camera design that images only specific types of desired objects while instantly erasing other types of objects from its images without any digital processing. In addition to data class-specific imaging, this AI-based camera design was used to create encryption cameras, which add an extra layer of security and privacy protection. Using AI-optimized diffractive layers, such an encryption camera optically performs a selected linear transformation only for the target objects of interest. Only those who have access to the decryption key (or, in this case, the inverse linear transformation) can recover the original image of the target objects. The information of the other unwanted objects, on the other hand, is irreversibly lost because the AI-designed camera all-optically erases them at the output. As a result, even when the decryption key is applied to the recorded images, it produces noise-like, unrecognizable features for other classes of undesired objects. This article continues to discuss how the AI-designed camera supports security and privacy. 

UCLA reports "AI-Designed Camera Only Records Objects of Interest While Being Blind to Others"