By David Crandall & Apu Kapadia
On July 26th SPICE co-sponsored the IEEE Conference on Computer Vision and Pattern Recognition. This workshop addresses the security and privacy implications of computer vision as it is integrated into the real world and cameras are installed in many consumer devices. Computer vision has been unleashed for everyday use, bringing vision applications mainstream. However, this success has also led to often overlooked consequences for information privacy and security. These concerns will only continue to grow as cameras are increasingly integrated into wearable and other pervasive computing devices in the next few years. Conversely, computer vision can also be used to enhance privacy and security, including assistive devices that help people with visual impairments assess their surroundings for privacy threats, website spoofing detection techniques that incorporate visual features, and forensic analysis tools that detect modifications to photos and videos. Some vision technologies present both challenges and opportunities: biometric-based authentication techniques like face recognition promise to enhance security, but biometric spoofing techniques are being designed to defeat them, while crowd surveillance techniques can be used to enhance public security, but could also be abused by repressive governments.
We need to better understand the potential threats of computer vision to people’s security and privacy, as well as the potential opportunities and applications for enhancing them. Otherwise, we risk creating technology that could completely undermine public safety and security, or that could face a backlash of public opinion if privacy is not sufficiently addressed. Unfortunately, the computer vision and privacy/security research communities are largely disjoint. The purpose of this workshop was to bring together researchers interested in the intersection between computer vision and security and privacy to explore these issues.
The workshop took place at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), one of the premiere conferences in the field. CVPR attracted approximately 5,000 attendees this year. Interest in our workshop was also quite high: the workshop featured 25 experts in computer vision and privacy/security on the programcommittee, our call for papers attracted 12 full papers and over 20 additional extended abstracts, and an audience of about 100 students, researchers, and practitioners. Papers included diverse topics at the intersection of computer vision, privacy, and security, from uncovering exploitable weaknesses of modern computer vision algorithms, to proposing hypotheses on the future role of computer vision in society, to developing new techniques for embedding and detecting image watermarks, among many others.
A major goal of the workshop was to help spark interest among young researchers in the important emerging intersection between privacy and security and computer vision. Since this is a new area, we anticipated that many students may not have faculty with project funding to support travel to the workshop. Thanks to support from the National Science Foundation, SPICE, and Pixm, Inc, we were able to sponsor about 15 student travel awards.