In recent years, biometrics have been used in a variety of IT systems. Many biometrics require contact with the device, but during the 2019 corona pandemic, contact with a device shared by multiple people was a risk of contact transmission. In addition, facial recognition, which does not require contact and is used in iPhones, requires a three-dimensional sensor in addition to the usual camera and is considered to be costly. We proposed an authentication method that uses an image of a bent finger as a contactless, highly accurate, and impersonation-resistant authentication method. When a human tries to bend only the middle finger intentionally, the other fingers bend as well unintentionally, and the authentication is performed using the fact that the shape of the bent fingers differs from person to person. We examined its authentication ability using machine learning. The machine learning method applied to this system were Google MediaPipe to detect the joints, and One Class SVM to identify the person himself/herself from others as anomaly detection. The results of the authentication experiment showed that the authentication accuracy was sufficient for a biometric authentication system.