Figure 8 - uploaded by R.Dhanapal Rd
Content may be subject to copyright.
Confusion chart for the classification of fake face image (class 0) and real face image(class 1) Recently a wide variety of feature selection algorithms (Alfeilat et al., 2019 ; Franti, P.,2018 ; Jingwei Too, 2020) are available
Source publication
Face is a very popular non-intrusive modality used in biometrics recognition. Spoofing attack is a method using which an illegitimate access is made via the face biometric system. Face anti-spoofing technique we implemented using static approaches are based on descriptors that are extracted using Local appearance-based methods for the face images o...
Contexts in source publication
Context 1
... can be considered as a future direction to increase the real face recognition rate to a decent value. Achieved results for the test samples are demonstrated using confusion chart in Figure 8. These new set of features further go on and improve the classification accuracy with respect to real face recognition rate. ...
Context 2
... can be considered as a future direction to increase the real face recognition rate to a decent value. Achieved results for the test samples are demonstrated using confusion chart in Figure 8. These new set of features further go on and improve the classification accuracy with respect to real face recognition rate. ...
Citations
... Major ones are texture-based, temporal-based, and frequency based. All the above approaches incorporated handcrafted texture features like LBP, HOG, and its variants followed by conventional classifiers like K-means, Support Vector Machines (SVM), or Neural networks to carry out the anti-spoofing task [6,7]. To distinguish between real faces and fraudulent ones, the temporal-based approaches [8] utilize facial motion patterns like eyes blinking or facial movements that use optical flow for face movement tracking. ...
... Dynamic procedures examine the images based on their temporal qualities while static approaches assess the images based on the spatial relationship in the image. Static techniques that utilize local binary patterns and their variants [6,12,13], Fourier analysis [7,14], Difference-of-Gaussian (DoG) [14], and Lambertian models [15]. Dynamic methods benefit from the temporal correlation between subsequent frames of video [16]. ...
... The authors also presented a cross-dataset evaluation to show the generalisation capability of the method. The authors of [18] combined different handcrafted features including LBP, GDP, GLTP, LDIP, LGBPHS, and LPQ. These extracted features were classified by using the K-NN classifier. ...
Face presentation attacks (PA) are a serious threat to face recognition (FR) applications. These attacks are easy to execute and difficult to detect. An attack can be carried out simply by presenting a video, photo, or mask to the camera. The literature shows that both modern, pre-trained, deep learning-based methods, and traditional hand-crafted, feature-engineered methods have been effective in detecting PAs. However, the question remains as to whether features learned in existing, deep neural networks sufficiently encompass traditional, low-level features in order to achieve optimal performance on PA detection tasks. In this paper, we present a simple feature-fusion method that integrates features extracted by using pre-trained, deep learning models with more traditional colour and texture features. Extensive experiments clearly show the benefit of enriching the feature space to improve detection rates by using three common public datasets, namely CASIA, Replay Attack, and SiW. This work opens future research to improve face presentation attack detection by exploring new characterizing features and fusion strategies.
... A recent work by Thippeswammy, Vinutha, and Dhanapal [14] proposed a method based on an ensemble of texture descriptors of local appearance-based methods. In their approach, Local Binary Pattern (LBP), Gradient Directional Pattern (GDP), Gradient Local Ternary Pattern (GLTP), Local Gabor Binary Pattern Histogram Sequence (LGBPHS), Local Directional Pattern (LDiP), and Local Phase Quantization (LPQ) descriptors were concatenated to form an ensemble feature vector representation of images. ...
Abstract— The applicability, popularity, and usage of face
recognition systems are undoubtedly increasing due to their
advantages of being very convenient, contactless, and non-intrusive when compared with other biometric systems such as
voice recognition and fingerprint. Unfortunately, it is the most
vulnerable to spoofing attacks as a registered user’s
photographs/videos can be gotten with ease through the internet
or simply capturing their face using a camera, even without the
consent or physical contact with him/her. Hence, the need for
the development of anti-spoofing measures against such attacks.
Several efforts have been in place in research for the
development of face anti-spoofing methods. In this study,
authors reviewed various anti-spoofing methods by several
authors by searching related papers using keywords from online
sources. This paper discusses several approaches to face spoof
detection, the different types of face-spoof attack, and provide a
taxonomy of the various anti-spoofing methods. The results are
presented in a tabular form, providing a comprehensive list of
publicly available face anti-spoof databases and a performance
comparison table of several approaches.
The significance of Face Recognition and Face Biometric Authentication systems has grown significantly in diverse security applications; nonetheless, the ability of the facial identification system to withstand an attack remains a significant concern. Research on non-invasive software-centric face spoofing detection systems have primarily focused on analyzing the luminance information present in face images, neglecting other components such as chroma and intensity which has the potential to be quite valuable for discriminating false faces from genuine ones. The proposed research work aims to utilize these components to find the differences or inconsistencies that could indicate the presence of a face spoofing attack. It is done by extracting and analyzing crucial information including energy, color distributions, and brightness of the color channels in an image. First, texture feature is extracted using Gray-Level Co-occurrence Matrix, GLCM. Second, the statistical descriptors features of RGB, HSV, YCbCr, CMYK, YIQ and YUV color channels are extracted to get the color distribution information. Subsequently, the integration of texture and color channel features results in the formation of a refined and enriched feature vector. At last, the feature vector is inputted into CNN architecture to classify the spoofed and genuine face images. The proposed method is assessed on NUAA Photograph Imposter dataset and has obtained a test accuracy of 99.88% in face spoof detection with HTER of 0.01%.
