FACE SPOOFING DETECTION USINGCOLOUR TEXTURE ANALYSIS

CONTENTS

• INTRODUCTION

• EXISTING METHOD

• PROPOSED METHOD

• ADVANTAGES

• APPLICATIONS

• CONCLUSION AND FUTURE SCOPE

• REFERENCES

INTRODUCTION

• Detection using colour texture analysis

• Information from luminanace and chrominance are collected

• Existing method focussed on analysis of luminanace

information of face images and discarding chroma component

• Observed fake image have lower image quality with lack of

high frequency information

Contd…

• Fake images are identified by analysing chroma component

than luminance

• Preliminary colour texture analysis approach is proposed

• Non intrusive software based detection focus on gray scale

images, discarding colour information

EXISTING METHOD

• Hardware based solution appoach

Surface reflectance properties are used

Thermal information are for detecting printed and replayed

video attacks

But are intrusive, expensive or impractical since

unconventional imaging devices are required

Contd…

• Challenge response approach

Specific action is choosed as challenge and actually performed

or not, is response

• Non intrusive approach

No user co-operation is required

Assessed by commonly used any database

Categorized into static and dynamic tech

Contd…

High resolution input image are required to extract fine details

Generation capabilities are not clear due to lack of training and

testing set

Colour local binary patterns descriptor is only used

PROPOSED METHOD

• Face spoofing attacks mostly performed by displaying using

prints, video displays or masks

• Detect by analysing texture and quality of captured gray scale

image

• Discriminating genuine faces from fake ones by insight image

into three colour spaces RGB, HSV, YCbCr

Contd…

• Performance of different facial colour texture representation is

compared to their gray scale

Contd…

• Similarity between LBP descriptions extracted from face 1 and

face 2 for printed and video attacks

• Similarity is measured using the chi-square distance

• Hx and Hy are two LBP histograms

• Chi-square distance between gray-scale LBP histograms of the

genuine face and the printed fake face is smaller than the one

between two genuine face images

Contd…

• Mean LBP histograms for both real and fake face images to

compute a Chi-square distance as

• Hx is the LBP histogram of test sample and Hr & Hf are the

reference histograms for real and fake faces

Contd…

• Score distributions of the real faces and spoofs in the gray-

scale and YCbCr colour space.

• Chi-square statistics of the real and fake face descriptions in

the gray-scale space and Y channel are overlapping

• Better separated in the chroma components of the YCbCr

space.

Contd…

Proposed face anti-spoofing approach

Contd…

• Face is detected, cropped and normalised into an M×N pixel

image

• Texture descriptions are extracted from each colour channel

• Resulting feature vectors are concatenated into an enhanced

feature vector to get an overall representation of the facial colour

texture

• Final feature vector is fed to a binary classifier

• Output score value describes whether there is a live person or a

fake one in front of the camera

Contd…

• Facial representations extracted from different colour spaces

using different texture descriptors can also be concatenated

• Colour space

Two other colour spaces, HSV and YCbCr, to explore the colour

texture information in addition to RGB

HSV colour space, hue and saturation dimensions define the

chrominance and while the value dimension corresponds to the

luminance

Contd…

YCbCr space separates the RGB components into Y, Cb and

Cr

• Texture Descriptors

Designed for gray- scale images can be applied on colour

images by combining the features extracted from different

colour channels

5 descriptors

Contd…

Local Binary Patterns (LBP)

• Binary code computed by thresholding

• Binary patterns are collected into histograms

Co-occurrence of Adjacent Local Binary Patterns (CoALBP)

• LBP discards spatial information

• To exploit the spatial relation between patterns

Contd…

Local Phase Quantization (LPQ)

• Deal with blurred image

• Phase information extracted by STFT to analyse neighbourhood

• Quantized and collected into histograms

Binarized Statistical Image Features (BSIF)

• Convolving the image with linear filter and binarizing filter

response

Contd…

Scale-Invariant Descriptor (SID)

• Image is first re-sampled densely enough on a log-polar grid,

rotations and scalings in the original image domain

• Fourier transform is applied on the re-sampled image,

invariance to both scale and rotation is achieved

ADVANTAGES

• Do not require any additional sensor

• Focused on both printed and replayed video attacks

• Good generalization ability

• Low computational complexity

• Fast response

• CTA features are more robust

APPLICATIONS

• Authentication system

• Registration purpose

• Mobile payment

• Unlocking system

• Security purpose

CONCLUSION AND FUTURE SCOPE

• Approach the problem of face anti-spoofing from the colour

texture analysis

• Colour image representations can used for describing the intrinsic

disparities in colour texture

• Facial colour texture representations studied by extracting

different local descriptors

• Improving generalization capabilities of colour texture analysis

based face spoofing detection

REFERENCES [1] Zinelabidine Boulkenafet, Jukka Komulainen, and Abdenour Hadid,”Face

Spoofing Detection Using Colour Texture Analysis”, IEEE Transactions On Information Forensics And Security, Vol. 11, No. 8, August 2016.

 [2] Y. Li, K. Xu, Q. Yan, Y. Li, and R. H. Deng, “Understanding OSN-based facial disclosure against face authentication systems,” in Proc. 9th ACM Symp. Inf., Comput. Commun. Secur. (ASIA CCS), 2014, pp. 413–424.

 [3] J. Li, Y. Wang, T. Tan, and A. K. Jain, “Live face detection based onthe analysis of Fourier spectra,” Proc. SPIE, vol. 5404, pp. 296–303,Aug. 2004.

 [4] X. Tan, Y. Li, J. Liu, and L. Jiang, “Face liveness detection from a single image with sparse low rank bilinear discriminative model,” in Proc. 11th Eur. Conf. Comput. Vis., VI (ECCV), 2010, pp. 504–517.

 [5] Z. Zhang, J. Yan, S. Liu, Z. Lei, D. Yi, and S. Z. Li, “A face antispoofing database with diverse attacks,” in Proc. 5th IAPR Int. Conf. Biometrics (ICB), Mar./Apr. 2012, pp. 26–31.