Master Assignment
FISWG AUTO – Automatic extraction and use of forensic facial features
Type: Master EE/CS
Period: TBD
Student: (Unassigned)
If you are interested please contact :
Description:
Forensic Facial Recognition - examiners compare crime scene images and reference images taken from a suspect and formulate a (descriptive) estimation of the strength of evidence that the images depict the same person. A judge can incorporate the strength of evidence in the verdict whether the suspect is considered guilty or not. The comparison protocol typically involves the assessment of (dis)similarities found during a morphological analysis; its details may vary between forensic organizations. Since the comparison process is to some extent subjective, insight into decision making and efforts to objectification are important. FISWG (www.fiswg.org), a scientific working group in which facial identification knowledge and experience is organized, has published recommendations for different levels of the comparison process. Their facial comparison list describes overall and detailed FISWG characteristic descriptors, that is, typical features of facial parts like the eyebrows, eyes, nose etc.
Several studies have been conducted on FISWG characteristic descriptors on images of various quality. These pre CNN studies mainly used manual annotation to extract features. It is interesting to see whether (a limited collection) of these characteristic descriptors can be extracted automatically and used for biometric purposes. Also, many studies have been performed on so called soft biometrics on parts of the face that predominantly use texture information.
Assignment
The assignment has three parts
- Prepare a dataset.
- Design and implement a system that can extract FISWG characteristics of a limited set of facial parts. Study the biometric performance of the characteristic descriptors, also in comparison to traditional texture based systems.
- Study the effect of biometric fusion, that is, when you combine the two approaches.