Technical specifications for the identification of synthetic human face images

SF/T 0148-2023

Introduction

This document is in accordance with the provisions of Part 1 of the Standardized Working Guidelines GB/T 1.1-2020: Structure and Rules for Drafting of Standardized Documents

Drafting.

Please note that some elements of this document may relate to patents. The issuer of this document does not have the responsibility to identify patents.

This document was submitted by the Institute of Forensic Sciences.

This document was sent back by the Information Centre of the Ministry of Justice.

Prepared by the Institute of Forensic Sciences, the University of Nakayama, Shenzhen University, the Gangnam School of Social Sciences, and the Shangping Judicial Accreditation in Shanghai

Centre.

Principal drafters of this document: Tsang Jinhua, Shigebe, Guo Hong, Yang Xue, Xin Wei, Cheng Jianhua, Lu Kai-ming, Li Iwa, Guangpao, Tana

Noon, Yang Jie, Mao Xiao, Ling Jing, Li Zhijun, Lou Wei, Li Bin, Yao Qing, Lu Mfun, Chu Tongtao. SF/T 0148-2023

Technical specifications for the identification of synthetic human face images

1 Scope

This document sets out the equipment, content and technology for the identification of synthetic human face images in audio-visual identification, synthetic human face images

Test methods, test steps and expert opinion.

This document applies to the visualization of synthetic human faces in the forensic field.

2 Normative references

The contents of the following documents constitute essential provisions of this document through their normative references. of which, the date of reference,

Only the version corresponding to that date applies to this document; date-free references, the latest version (including all changes) apply to this document

Documentation.

SF/T 0119 General Guidelines for the Identification of Audio Images

SF/T 0120 Technical specifications for authentication of sound recordings

SF/T 0122 Technical instruction for speech homogenization

SF/T 0123 Technical specifications for video authenticity

SF/T 0124 Technical specifications for video process analysis

SF/T 0125 Code of Technology for Identification of People

SF/T 0153 Technical specifications for photo authenticity

3 Terminology and definitions

SF/T 0119, SF/T 0120, SF/T 0122, SF/T 0123, SF/T 0124, SF/T 0125, SF/T 0153 and

The following terms and definitions apply in this document.

Synthetic face image

Images of human faces are processed later in the image.

Checklist face image

Synthetic face image identification is required.

Inspection image

(b) Recording images with image (3.2) of the person's face.

Reviewed voice

The voice message of the inspecting body is recorded in a video carrier.

Sample human face image known

Images of human faces for comparison and comparison.

Sample voice known

Voice of the talking person for comparison and comparison.

facial properties operation

Synthetic modifications to individual or more facial components or properties of human face images. SF/T 0148-2023

Example: Synthetic modification of colour, hair and gender.

Facial expression operation

Synthetic modification of the facial expression of a human face image.

Example: Synthetic modifications to facial muscle changes and lip movements, etc.

Face replacement

Replacement modifications to facial areas of human face images.

Generated facial image

Images of human faces synthesized through artificial intelligence algorithms with no corresponding target identity.

Synthetic image recognition

Integrated application of human identification, speech homogeneity through technical means such as visual recognition, imaging analysis, voice analysis, signal statistical analysis

Technology such as identification, authentication of image authenticity, authentication of audio recording authenticity, image of the human face of inspection recorded in vectors such as videos or photographs (3.2)

Whether it is scientifically judged through facial properties, facial expressions, facial replacements, or whether it is the generation of a human face.

4. Instrumentation

Equipment and tools for the identification of synthetic human face images include:

(a) Image capture and backup equipment: including image distribution equipment, image capture equipment, writing protection tools, completeness validation calculations Workers

Tools, complete backup and format conversion tools, etc.;

(b) Image analysis equipment: including image analysis tools, sound spectrum analysis tools, signal analysis tools, image processing tools and audio processing

Tools, etc.;

(d) Specialized image authenticity identification system.

5. Test content and test technology

Test Contents

Synthetic face image identification includes the following tests:

a) the whole area of the head and face of the image of the human face;

(b) Face components and properties of human face images;

(d) Head movement of human face images;

e) facial expression and facial muscle changes in the image of the human face;

(f) The movement of the lips and eyes of the image of the human face;

(g) Relevance and consistency of behaviour and sound in the head and face of images examined in video carriers.

Test techniques

Synthetic face image identification includes, but is not limited to, the following test techniques:

(a) Image authenticity identification techniques: an analysis of image characteristics from the perspective of image analysis;

(b) Audio-visual authentication techniques: speech profiling from the perspective of voice analysis;

(c) Human identification techniques: checking the identity of the person in the image of the person ' s face from the point of view of his or her identity;

(d) Voice homogeneity identification technique: the identification of the voice of the inspection material in the image of the subject's face from the perspective of voice homogeneity

Inspection;

(e) Mode-recognition techniques: review of the speech characteristics of the image or the voice of the scrutinizing material from the perspective of the statistical analysis of signals

Check.

6 Test method SF/T 0148-2023

Image characterization

Image characterizations include the following main elements:

(a) The light and color distribution of the profile;

(b) The ratio of visualization of images of the subject ' s face;

(d) The visual scene and perspective of the image of the subject;

(e) The alteration and modification of images of the person ' s face;

(f) Irregular specks, deformations and misappropriations in the image of the target person;

(g) Distribution of pixels at the edge of the surface area of the profile;

(h) Checking the image quality of the image;

(i) Activation of images of the human face examined in video carriers, such as head movement, facial expression, facial muscle change, eye movement and

Lip movement, etc.;

(j) The correlation between the action expression and the sound of the image in the video carrier;

k) Consistency and continuity of image characteristics of the human face in video carriers.

Voice characterization

Voice characterization includes the following main elements:

(a) Consistency and homogeneity of speech;

(b) Relevance and logic of the speech expression;

(d) Examination of the nature of the speech and its compatibility with the subject matter;

(e) Sources and changes of the background sound;

(f) Continuity and consistency of the narrative;

(g) Overlap between background sound and censorship;

(h) Examination of the beginning and end of the speech;

(i) Examination of spectrum changes in voice;

(j) Spectrum mutations of signals;

(k) Repetition of signals for sound.

Identification

6.3.1 Identification of identity in the image of the target person and the image of the sample person, in accordance with SF/T 0125

Done.

6.3.2 Voice homogeneity identification of sound and sample speech in the profile of the subject, according to SF/T 0122

Relevant provisions are made.

Action Behaviour Test

6.4.1 Test and analysis of the behaviour of the profile of the subject, in accordance with the relevant provisions of SF/T 0124.

6.4.2 Consistency and reasonableness tests are conducted for head and face movements of the profile of the subject.

6.4.3 Consistency and reasonableness tests are performed on the behaviour of the facial components of the profile.

6.4.4 Where applicable, correlation and consistency tests are performed on the physical behaviour of the profile and the speech of the inspection material.

Signal statistical analysis

6.5.1 Statistical analysis of background noise distribution of inspection images.

6.5.2 Where applicable, statistical analysis of background noise distribution of detected speech.

6.5.3 Test identification of synthetic human face images using alternative modes of identification and machine learning.

7 Test steps

Information SF/T 0148-2023

7.1.1 Knowledge of the process or origin of the inspection material and samples.

7.1.2 Identification of the content of the images.

Image Collection

7.2.1 For simulated images, analogue images are converted using appropriate collection methods, such as scanning, photo and play collection

Digital images.

7.2.2 For digital images, images are collected through document reproduction. If the original image is involved, it is advisable to reproduce it by writing protection.

7.2.3 For voices in digital video, it is advisable to turn the sound in the video in a non-destructive manner while maintaining the quality of the sound

For digital recording.

7.2.4 Conversion of collected images into file formats supported by the inspection system through the format conversion tool, as necessary.

7.2.5 Unique identification of collected images and recordings and calculation of completeness verification values.

Image Processing

7.3.1 When applicable, detection images may be reduced or enhanced for processing.

7.3.2 Image processing shall not alter the essence of the image data.

Image Check

7.4.1 General requirements

The appropriate methods in 7.4.2 to 7.4.7 are selected for testing and analysis, depending on the circumstances of the inspection images.

7.4.2 Synthetic image characterization

Synthetic inspection of the image profile based on the image characterization method (6.1)

7.4.3 Synthetic speech characterization

Synthetic testing of the profile of the spotter based on the speech characterization method (6.2).

7.4.4 Synthetic identity tests

Synthetic testing of the profile of the subject is carried out on the basis of identification methods (6.3).

7.4.5 Synthetic action behaviour tests

Synthetic inspection of the profile of the examiner based on the action behaviour test method (6.4).

7.4.6 Synthetic signal statistical analysis tests

Synthetic testing of the image of the spotter's face based on signal statistical analysis methods (6.5).

7.4.7 Image authenticity test

Under the relevant provisions of SF/T 0153 and SF/T 0123.

Comprehensive assessment

In accordance with the accreditation requirements, the various phenomena detected in the examination were examined, taking into account the circumstances of the case and the presentation of the image of the subject ' s face.

As a result, a system analysis was carried out to determine whether the inspection video had been synthesized and an assessment made accordingly.

8 Validation

Classification of assessments

The identification of synthetic face images was divided into six categories:

(a) Synthetic processing;

b) has not been synthesized;

d) Predisposition to synthetic treatment; SF/T 0148-2023

(e) The tendency to be unsynthetic;

f) It is not possible to determine whether they have been synthesized.

Basis of judgement and presentation

8.2.1 Synthesis

8.2.1.1 Basis of judgement: Discrepancies in the image of the subject ' s face were found and analysed for synthetic processing.

8.2.1.2 Statement of the opinion: The image of the subject ' s face has been synthesized.

8.2.2 Not consolidated

8.2.2.1 Basis of judgement: No anomalies were found in the profile of the subject ' s face and no synthesis that could not be discovered through available technical means was analysed

Deal with possibilities.

8.2.2.2 Representation of an expert opinion: The image of the subject's face has not been synthesized.

8.2.3 No synthesis found

8.2.3.1 Basis of judgement: No anomalies in the profile of the subject ' s face were detected or could reasonably be explained, but not yet fully scheduled

Except for the possibility of synthetic treatment traces that are difficult to detect based on available technical means.

8.2.3.2 Statement of the opinion: No image of the subject's face was found to have been synthesized.

8.2.4 Prefer to synthetic treatment

8.2.4.1 Basis of judgement: Discrepancies in the image of the subject ' s face were detected and the potential for these anomalies to be formed as synthetics was analysed.

8.2.4.2 Representation of the expert opinion: There is a tendency to assume that the image of the subject is synthesized.

8.2.5 Propensiveness without synthesis

8.2.5.1 Basis of judgement: Anomalous images of the subject ' s face are found, but these anomalies can essentially be explained and analysed through synthesis

It's not likely.

8.2.5.2 Representation of the expert opinion: There is a tendency to believe that the profile of the subject has not been synthesized.

8.2.6 Uncertainty as to whether it has been synthesized

8.2.6.1 Basis of judgement: The profile of the profile is abnormal, but cannot be judged by its nature or cause; or the profile of the profile is not available

Conditions for accreditation.

8.2.6.2 Representation of the expert opinion: it is not possible to determine whether the image of the subject was synthesized.

Any questions, group feedback! Expression 26@2x