Abstract:
Conventional biometric systems do not possess the capability to detect whether a biometric image is
acquired from a live subject or an artificial representation of his identity. This allows impostors to use
different methods to fake the identities of legitimate users and compromise the security of biometric
authentication systems. This paper proposed a texture-based anti-spoofing technique known as
concatenated rotation invariant uniform local binary pattern, which uses textural properties to
discriminate between images (face and iris) captured directly from live subjects and those obtained
from secondary sources such as photographs or video images. The proposed approach extracts
uniform local binary pattern features from an image at different scales and resolutions. The extracted
features are further concatenated to obtain a composite feature representation of the image. The
accuracy of proposed method is evaluated using face images from Nanjing University of Aeronautics
and Astronautics (NUAA) normalized dataset and iris images from Audio, Temporal signals, Vision
and Speech (ATVS) fake iris database. The programming environment used to implement the
proposed technique is MATLAB 2014. The MATLAB environment provides the tools/utilities for
creating the programs which implements the various tasks in the spoof detection system. Experimental
results suggest that the proposed approach is capable of distinguishing genuine face and iris images
from fake representations of the same image. The results also show the technique has better
recognition accuracy and higher textural discriminative power for iris than it does for face. This is
largely due to the fact that iris exhibits low intra-class variation and high inter-class distance; while
face has high intra-class variation and low inter-class distance. The suitability of the proposed
technique is not limited to only face and iris biometric data. The technique can be applied to any
biometric modality whose textural features can be extracted. Examples include retina, palm, knuckle,
fingerprint, lip and ear. We only used face and iris samples to verify the proposed method
