Typologie des traces Cours 6

Typologie des Traces - Reconnaissance Faciale

• Reconnaissance faciale (1er moyen): Focuses on the general facial structure, including elements like the shape of the face, proportions, eye color, skin tone, hair, birthmarks, scars, and freckles.
• Context: Challenges in police work include low image quality, poor angles, and individuals avoiding cameras, making reliable identification difficult. Three main approaches exist:

Approches de Reconnaissance Faciale

• Approche 1 (1v1 Verification): Comparing an unknown person's picture to a known reference image. Used for forensic expertise, border control, and access control (e.g., airport, company).

• Approche 2 (1vN Identification): Comparing an unknown image to a database. Used for detecting series of offenses, identifying suspects, and finding duplicates in databases (e.g., driving licenses). This process is automated due to large databases (like driving license databases.)

• Approche 3 (Supervision 1v0): Tracking a person in a video sequence. Useful for investigations of a person's location and activities. Images are not always of high quality (e.g., black and white, varying resolutions).

• Perception of facial recognition: Society's perception of facial recognition is often influenced by media and TV shows, leading to unrealistic expectations of its accuracy and effectiveness. The public often misunderstands its limitations. Error rates can be high compared to expert human judgment or even non-experts, especially when there is no specific protocol or conditions.

Fonction Élémentaire

• Trace analysis: Determining the source of traces, relating cases, and identifying the person/entity behind a trace. Visages are crucial for this, allowing for reconstruction of events and people.

Historique de la Reconnaissance Faciale

• Physiognomonie: Historical approach using facial features to infer character and personality (e.g., Cesare Lombroso).
• Photographie (Daguerréotype/Bertillonnage): These methods involved taking photos and creating standardized facial descriptions, marking a transition from non-standardized to standardized practices, making facial recognition more scientific.

Approches Comparatives Traditionnelles

• Analyse holistique: Considers the whole face. Quick but less reliable.
• Analyse morphologique: Focuses on facial characteristics using descriptive language. Less objective though, and subject to expert variation and interpretation, affecting the accuracy, especially when there is no agreed upon criteria or set guidelines.
• Analyse anthropométrique: Measures specific points on the face, offering more objectivity by using precise point-to-point measurements, a process potentially less prone to biases.

Analyse par Superposition d'Images

• Juxtaposition: Comparing multiple images of the same face to check for similarities or differences in the same conditions. Useful for identification.
• Approaches Actuelles: Combining traditional with automated methods, using automatic systems to generate potential candidates of suspects, improving efficiency and the speed of identification but with limitations and possible biases.

Technologie Émergente : Deep Learning

• Systèmes entrainés: Large datasets for facial recognition, showing promising results but potentially lacking transparency regarding their methods and potentially biased outcomes (ex: bias based on ethnicity).
• Performances variables: Performance can vary depending on the training data's characteristics, leading to mislabeling of some people/groups. Deep Learning systems can be susceptible to errors when images are of poor quality, from different angles or lighting conditions, or when images aren't well aligned in the required format. These systems can also lack transparency, meaning it is not always completely obvious how they make a determination.

Débats Sociétaux

• Droits et libertés: Legal debate on the use of automatic facial recognition, including concerns about privacy, government surveillance, and the right to be safe.
• Algorithme bias: Issues with bias in algorithms, affecting different ethnic or racial groups or people, potentially increasing risk of unfair or inaccurate identification.
• Sécurité face à la collecte des données: Concerns over the collection, storage, and use of facial recognition data, including the potential for misuse, and limitations in maintaining security or protecting individuals from identity theft or inappropriate data access.
• Proactiveness of identification: Debate on the ethics of proactively identifying individuals—e.g., at public places like stadiums or streets, raising concerns on intrusion of privacy, personal boundaries, legal repercussions, and potential misuse.