Facial Recognition for Forensic Video Analysis
Facial recognition experts are highly trained Forensic Video and Image analysts who specialize in the fields of human diversity, anthropology and forensic analysis. Even though we recognize faces every day, but to analyze and replicate this same behavior of a human brain through scientific procedures and then to testify in court that a suspect is a criminal, is not as easy as we might think. This requires the expertise of a facial recognition forensic analyst. They use techniques such as Morphological analysis, Anthropometric, Photographic Superimposition and 3D–3D comparison to recognize faces in a surveillance camera shoot.
With the growing use of video surveillance cameras, the need for facial recognition in forensic video analysis is also growing. There are cameras fitted almost everywhere, on roads, malls, parking lots, lobby, homes, offices, airports, entry and exit doors of every other building. Usage of facial recognition for forensic analysis dates back to the end of 19th century, when Alphonse Bertillon, a French police officer, first used anthropometry for the purpose of facial recognition. In literal terms anthropometry is the study of human body measurements especially on a comparative basis. However forensic anthropometry for the purpose of facial recognition, involves a careful comparison of morphological characteristics of some peculiar anatomical parts (such as eye, nose, mouth, and ear) of the head.
As explained above facial recognition is an essential and difficult aspect of forensic video analysis. What make it even more difficult are limitations such as disguises to hide identity and poor quality of surveillance camera footage. Though special forensic enhancement software are available that enable demultiplexing, frame averaging, duplication, video level adjustment, magnification, highlighting, and obscuring of multiple subjects and specific areas. But even post enhancement, a forensic specialist, preferably an expert in the fields of biology/anthropology/human-diversity, is required to recognize faces, especially when the original quality of surveillance videos or images is very low.
Face Recognition methods for Forensic video analysis
Forensic facial identification involves comparison of two or more faces to recognize the facial image in question and determine the true facial identity. Amongst the traditional approaches used for face recognition are 3 well known techniques. These are Morphological analysis, Anthropometric and Photographic Superimposition.
- Morphological Analysis – This is a very old approach of facial recognition. It involves comparison and identification based on individual features of a face. Several classifications of different forms of facial regions and types exist. These classifications aim at meticulously classifying different facial traits, such as, facial outline shapes, hairline shapes, mouth, nose, etc. Some classifications can consider up to 40 facial traits. They are designed to promote a consistent, systematic, and scientifically comparable evaluation of facial features as well as making individual variations and population differences emerge from a seemingly unremarkable visage. It can guide the expert in making a decision about the final match.
- Anthropometric approach – This approach is described as the quantification of physiological proportions between facial traits, dimensions, ratios and angles to measure specific characteristics of a face. It is generally used for facial comparison of similar face orientations. Even slight differences in orientation, facial expression, lighting conditions, camera distortions, camera positioning and aging can impact the end results and will require proper video enhancement. The method proposed by Alphonse Bertillon was based on anthropometric measurements. He developed a taxonomy, called as portrait parle or spoken portrait, which described the physiological features of the head, nose, forehead and ears. This combination of anthropometric measurements and the spoken portrait was called Bertillonage and was very soon adopted by judicial systems of that time.
- Photographic Superimposition – This technique involves comparison of facial traits by superimposing them on each other. Forensic experts take pictures or videos of the suspect with closest possible head orientation of the criminal, and then compare facial characteristics through superimposition and numerical analysis of two images. Favorable recording conditions are essential and this technique is usually impractical, especially if long time has elapsed between the occurrence of crime and the facial recognition or forensic analysis.
- More modern approach to facial recognition techniques involves comparison in 2D or 3D structures. With technology and software advancements the shift is also towards automating the identification approaches. Forensic experts are using quantitative identification procedures like 2D-2D, 3D-2D and 3D-3D comparison techniques to identify the criminals. These are more reliable and accurate techniques.