Matrix-Regularized One-Class Multiple Kernel Learning for Unseen Face Presentation Attack Detection

Arashloo, Shervin Rahimzadeh

doi:10.1109/tifs.2021.3111766

Cited by 15 publications

(41 citation statements)

References 77 publications

(103 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The work in [8] extends the one-class kernel Fisher null method to an MKL setting by imposing a matrix p,q -norm constraint (p, q ≥ 1) on kernel weights. Assuming G base kernels K g , g = 1, .…”

Section: Preliminariesmentioning

confidence: 99%

“…In contrary, the more commonly used vector p -norm regularisation [29] lacks an explicit mechanism to capture inter-kernel interactions. As such, the matrix-norm regularisation has been observed to outperform its vector-norm counterpart in different settings [8], [37], [34], [38]. Note that the vector-norm constraint may be considered as a special case of the matrix-norm constraint as setting p = q would reduce the matrix-norm into a vector-norm.…”

Section: Preliminariesmentioning

confidence: 99%

“…In this respect, when the test-phase attacks are fundamentally different from those observed during the training phase, the performance of the classical two-class closed-set systems may be seriously compromised. Due to its significant practical importance, detection of unseen presentation attacks has received considerable attention both in face [6], [7], [8], [9], [10], [11], [12], [13] as well as in other biometric modalities [14], [15], [16], [17], [18].…”

Section: Introductionmentioning

confidence: 99%

“…First, the MKL problem is formulated in the presence of an p -norm constraint on kernel weights. More recent studies have demonstrated that a more general matrix p,q -norm regularisation on kernel weights may yield better performance in different applications settings [8], [37], [34], [38]. Second, the localised MKL method in [36] decouples the learning tasks into a set of disjoint MKL problems each associated with a distinct region in the data.…”

Section: Introductionmentioning

confidence: 99%

“…In this paper, we study the locality information for unknown face PAD in the context of ensemble techniques and try to capture and use the inherent local structure in bona fide samples for improved detection performance. More specifically, drawing on the one-class MKL method of [8], we formulate a local multiple kernel learning one-class method addressing the aforementioned limitations of the existing localised MKL approaches and illustrate the utility of the proposed technique in detecting unknown/unseen face presentation attacks. In particular, in the proposed method, we regularise kernel weights using a joint matrix-norm constraint as compared with the existing local MKL methods using disjoint vector p -norm regularisation.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Unknown Face Presentation Attack Detection via Localised Learning of Multiple Kernels

Arashloo¹

2022

Preprint

Self Cite

View full text Add to dashboard Cite

The paper studies face spoofing, a.k.a. presentation attack detection (PAD) in the demanding scenarios of unknown types of attack. While earlier studies have revealed the benefits of ensemble methods, and in particular, a multiple kernel learning approach to the problem, one limitation of such techniques is that they typically treat the entire observation space similarly and ignore any variability and local structure inherent to the data. This work studies this aspect of the face presentation attack detection problem in relation to multiple kernel learning in a oneclass setting to benefit from intrinsic local structure in bona fide face samples. More concretely, inspired by the success of the one-class Fisher null formalism, we formulate a convex localised multiple kernel learning algorithm by imposing a joint matrixnorm constraint on the collection of local kernel weights and infer locally adaptive weights for zero-shot one-class unseen attack detection.We present a theoretical study of the proposed localised MKL algorithm using Rademacher complexities to characterise its generalisation capability and demonstrate the advantages of the proposed technique over some other options. An assessment of the proposed approach on general object image datasets illustrates its efficacy for abnormality and novelty detection while the results of the experiments on face PAD datasets verifies its potential in detecting unknown/unseen face presentation attacks.

show abstract

Section: Preliminariesmentioning

confidence: 99%

Section: Preliminariesmentioning

confidence: 99%