2018 25th IEEE International Conference on Image Processing (ICIP) 2018
DOI: 10.1109/icip.2018.8451291
|View full text |Cite
|
Sign up to set email alerts
|

Benchmarking Binarisation Schemes for Deep Face Templates

Abstract: Feature vectors extracted from biometric characteristics are often represented using floating point values. It is, however, more appealing to store and compare feature vectors in a binary representation, since it generally requires less storage and facilitates efficient comparators which utilise intrinsic bit operations. Furthermore, the binary representations are very often necessary for some specific application scenarios, e.g. template protection and indexing. In recent years, usage of deep neural networks … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

1
23
0

Year Published

2019
2019
2023
2023

Publication Types

Select...
5
2
1

Relationship

0
8

Authors

Journals

citations
Cited by 25 publications
(24 citation statements)
references
References 18 publications
1
23
0
Order By: Relevance
“…Also, the BFV scheme allows the use of the packing or batching technique [61] which encrypts multiple values into a single ciphertext (i.e., facilitates operations on vectors component-wise) and hence performs computations by using the SIMD (Single Instruction Multiple Data) primitives [62], [63]. Keeping in mind that the BFV schemes operate on integer values [44], face representations are firstly quantisised following the equal-width quantile strategy [64] which assigns for each float-value component in the vector an integer value.…”
Section: Template Encryptionmentioning
confidence: 99%
“…Also, the BFV scheme allows the use of the packing or batching technique [61] which encrypts multiple values into a single ciphertext (i.e., facilitates operations on vectors component-wise) and hence performs computations by using the SIMD (Single Instruction Multiple Data) primitives [62], [63]. Keeping in mind that the BFV schemes operate on integer values [44], face representations are firstly quantisised following the equal-width quantile strategy [64] which assigns for each float-value component in the vector an integer value.…”
Section: Template Encryptionmentioning
confidence: 99%
“…Two main classes of such approaches are feature transformation, with the aim to reduce the computational cost of individual template comparisons (see e.g. [25]), and preselection, with the aim of search space reduction, i.e. the number of necessary template comparisons (see e.g.…”
Section: A Computational Workload Reductionmentioning
confidence: 99%
“…Such comparators can take advantage of the more efficient bitwise operators, thereby reducing the computational workload. An illustrative example can be seen in [144] (and a simpler one in [143]), where various bit allocation schemes for float-based feature vectors generated by neural network-based systems are benchmarked. In [142], a new representation is extracted from minutiae points, which can be further binarised to accelerate the biometric template comparisons.…”
Section: Binarisationmentioning
confidence: 99%