Liveness Detection (Anti-Spoofing) Technology

The facial recognition system requires both facial recognition to verify "whether it matches the registered individual" and spoof detection to confirm "whether the person in front of the camera is a real human and not a photo or screen." We are pleased to announce that our company has started offering a facial recognition SDK as an add-on to the "spoof detection SDK" used in major medical institutions and in entry/exit management and attendance management settings. ■ What you can do with the set - By simply passing the same camera feed to both SDKs, you can integrate both identity verification and biometric detection. - In our validation, we were able to detect all face photos and smartphone screens that would pass through facial recognition alone when combined with spoof detection. - Facial verification accuracy: 99.37% (public benchmark LFW, in-house evaluation) - Compatible with Windows / Linux / iOS / Android We offer a one-month free trial version. For more details, please refer to the news release in the related links.

Swallow Incubate Co., Ltd. has started providing the latest version 2.4.0 of its "Impersonation Detection SDK," which detects "impersonation" using photos, smartphone screens, and printed materials during face recognition and identity verification, starting September 29, 2026. In this version, the core fake detection models (for landscape and portrait images) have been updated to enhance generalization performance, allowing for stable detection against unknown shooting environments and unknown attack patterns. Evaluation using approximately 1 million frames of in-house verification data has confirmed a detection accuracy of 99.9% (in-house evaluation). The detection engine for the "Impersonation Detection API" available on the cloud (AWS Marketplace) has also been updated to version 2.4.0. There are no changes to the pricing plans. A trial version of the SDK (Windows / Linux / iOS / Android) is available for free for one month.

Swallow Incubate Co., Ltd. has published a series of 10 technical articles on the determination of impersonation (liveness detection) in facial recognition systems, accompanied by implementation code and actual measurement values, on the "Swallow Incubate Technical Blog." All articles are available for free. Facial recognition is a technology that determines "who someone is," but it does not assess "whether that face is alive." When a standard facial recognition system is implemented using OpenCV and a printed photo of the person is presented, it can pass as the actual person. This series is structured into four parts for those responsible for the design, implementation, and procurement of facial recognition systems: 1. Facial recognition alone cannot prevent impersonation (actual measurements of implementation and photo-based bypass). 2. Types of attacks and how to interpret accuracy (differences between FAR/FRR and APCER/BPCER; limitations of three simple liveness methods). 3. Actual measurement information necessary for consideration of implementation (design for on-premises, closed networks, and embedded systems; face detection at 13.5 fps on Raspberry Pi 5; SDK integration in 23 lines of C++/22 lines of C#; retrofitting to existing systems). 4. A checklist of 12 design items. Reproduction procedures for attacks are not included. For details, please refer to the related links.

Swallow Incubate Co., Ltd. made its Liveness Detection API generally available on July 28, 2026 through AWS Marketplace, serving customers in Japan and worldwide. The API delivers over the cloud the same fake-detection engine as our Liveness Detection SDK, which is deployed at more than 40 sites across Japan. No SDK integration is needed: a single REST API call adds a live-or-fake decision to an existing face recognition or eKYC (online identity verification) flow. Liveness detection identifies presentation attacks - attempts to defeat face recognition by presenting a photograph, a smartphone screen or a printed image to the camera. The technology comprises fake detection, which decides from a single still frame, and action detection, which asks the user to perform randomized actions such as turning the head or blinking. This API provides fake detection, which requires only one still image. As of the end of June 2026, our liveness detection technology is deployed at more than 40 sites across Japan, in applications including authentication solutions for major medical institutions, access control terminals and attendance management terminals. Six plans, including a free tier of 50 calls per month, are available on AWS Marketplace.