The Anatomy of Recognizable Facial Features
This blog explores the connections between facial image comparison and the facial feature prediction practices that form the basis of facial approximation. Facial approximation is the practice of making an estimated recreation of an individual’s face from their skeletal remains using anatomical knowledge of the human face, head and neck. Both fields of facial imaging work towards a similar goal – identification, albeit in different forms. In addition to this, the analytical focal points in both facial image comparison and facial approximation, as well as the main components needed for familiar facial recognition, are the same, these being the primary features of the face.
The most recognizable facial features
Eyes, nose, mouth, ears. These four features are the first aspects of the face that we take notice of and register. The natural variation of facial features, the ways in which they are proportioned and configured in relation to each other, and the frequency and typicality of particular features and variance across populations all contribute to their recognizability when perceiving the whole of the face.
Humans typically excel at recognizing familiar faces, primarily based on perceiving internal facial features, and recognition is the driving investigative purpose behind the creation of facial approximations. We are even able to recognize people we know from low quality images. However, we are not as innately accurate at comparing or matching faces that are unfamiliar to us. When comparing facial images in forensic casework, the faces looked at are not familiar to the examiner. For this reason, facial examiners rely on a step-by-step morphological analysis of key internal facial features (the eyes, nose, mouth and ears), secondary aspects of the face (such as the hairline, cheeks, forehead, brows, chin, and jawline), and distinctive characteristics (like scars, creases, and marks) to identify similarities and dissimilarities between subjects in different images.
In facial image comparison, a subject’s four primary facial features are described and assessed based on sub-feature detail. For the eyes this can include such details as intercanthal and interpupillary distance, the appearance of the eyelids, the canthi, sclera, irises, and any noticeable asymmetry. Details of the nose include the root, base, body, and tip, as well as the appearance of the nostrils, alae, and columella.
The mouth demonstrates minute sub-feature detail such as lip creases, the vermilion border, and the cupid’s bow, with research showing that the height and width of the upper and lower lips are highly variable by sex and population affinity. The ear is abundant with distinctive structural detail, from fossae to tubercles, the shape of the helix, tragus, and antitragus, and the curve of the antihelix and the lobe. Research has demonstrated that the distinctive sub-features of the ear are highly accurate for facial comparison purposes, even for images of children. Ears are also finding growing utility in biometric facial image recognition and identification, due to their many specific sub-feature details which typically remain stable over time (with the shape and details of the ears developing in childhood and undergoing change in old age).
The anatomy of the features you see
The shapes, contours, and processes of our cranium and mandible are largely responsible for the appearance of facial features. It’s easy to see how the angle and definition of the frontal bone creates the slope of our forehead and the ridges of our brows, or how the edges and angle of the mandible translate to the definition of our chin and jawline. But in addition to this, the orbits of our eyes, the form of our nasal bones, the height and positioning of our teeth, the proportions of our face, and all of their minute bony details, all contribute to the individual physical expression of our facial features.
As with facial image comparison, in facial approximation, the sub-feature details are the focus of feature prediction methodology. Facial feature prediction has been a required standard practice for producing facial approximations since at least the mid-20th century, with facial soft tissue depth measurement being a founding component of the first reconstructions produced in the 1800s.
Size and distance measurements of bony features and fissures, and cranial landmarks, form the basis of facial feature prediction formulae that allow practitioners to estimate the appearance of particular sub-feature details. For example, the dimensions of the orbit are used to estimate the projection/prominence of the eyeball (which in turn affects the shape of the eyelids), the anterior position of the pupil and iris, and the locations of the medial and lateral canthus. Likewise, the shape, size, and angling of the nasal bones in relation to the anterior nasal spine can be used to predict nasal projection and the shape and direction of the nasal tip, while the width of the nasal cavity can be used to estimate the width of the nose and position of the alae. The height and gnathism of the teeth will influence the form of the lips, with some available formulae including estimation of the measurements of the cupid’s bow. Other smaller soft tissue sub-details of the mouth, however, cannot necessarily be predicted based off the bone. A general rule is that the width of the mouth – calculated based on the distance between the maxillary canine teeth – should align with the medial edges of the irises.
Many details on the skull are consequential when it comes to the expression of facial soft tissue. A bifurcated anterior nasal spine (the anterior nasal spine being located behind the columella of the nose) indicates a bifid nasal tip, while having prominent lacrimal crests (just inside the orbital rim beside the nose) with higher, rounder orbits may be a predictor that an individual will have epicanthic folds with their eyelids. These resulting sub-feature details are easily observable for drawing comparisons between sets of images and memories of familiar people.
Though the details of the ears provide many potential points of comparison for facial examiners, there is a lack of data on the relationship between the bony components of the ear region and the expressed soft tissue of the ears. This makes it very challenging for facial approximation practitioners to reliably estimate the appearance of the ear from the skull alone. There is debate about whether the size of the mastoid process (the large bony projection behind the external auditory meatus, or ear canal) influences the size of the ears, while some limited research suggests a correlation between the prominence of the supramastoid crest (of the temporal bone, above the external auditory meatus) and earlobe attachment. It is also possible that ear shape and formation are simply primarily determined by genetic, epigenetic, and embryonic factors for soft tissue expression, to the extent that the phenotype of the underlying bone ultimately plays a lesser role in how our ears end up looking.
Accuracy is key
Testing and method validation are vital to developing robust and scientifically-sound prediction techniques that can accurately estimate the dimensions and appearance of facial features. As discussed in our previous blog post, the lack of validation for many forensic methods is a well-known issue, and facial feature prediction protocols are unfortunately not an exception. Luckily, there have been large strides made in the past couple of decades to address this, with more research being conducted and literature published to validate protocols and develop regression formulae to more accurately estimate feature measurements based on more demographically diverse samples. Still, many standards often applied by practitioners are based on traditional suppositions and assumptions that either have not yet been subjected to rigorous scientific testing, or are not supported by new data, particularly when tested outside of their original study population.
Our ability to recognize faces is significantly affected by the positioning of the eyes alone, with some literature positing that the recognizability of a facial approximation can be affected by an estimation error as small as a couple of millimeters off from where the person’s eyes are supposed to be in life. Recognition studies have demonstrated that the nose area is also one of the first regions of the face that viewers fixate upon when studying a face to recognize it. Evaluations of commonly applied mouth estimation techniques have yielded inaccuracies, raising concerns as to how often facial approximations are created using over- or under-measured mouth widths that may impact their recognizability. These factors all highlight how important it is that the feature estimation techniques we apply to approximate faces be as accurate as possible in order to maximize the chance that the individual’s look is captured and recognized by family and/or friends.
In terms of developing facial feature prediction protocols, the ears are the most under-researched of the four primary facial features. Early data collected in cadaver studies in the 19th and 20th centuries formed the basis for many long-held preconceptions for ear estimation, however, more recently published methods for ear prediction have failed to be supported by subsequent test studies. As a result, existing ear estimation formulae rely on data collected on aging trends and the premise that the ears tend to lengthen as we get older, while the sub-feature details remain relatively stable. These formulae also factor in age and sex to approximate the basic dimensions (length/height and width/breadth) of the ear. Additional studies on the hard-to-soft tissue relationship will be crucial to developing our understanding of how to predict ear expression from the skull.
Another necessary direction for validation studies will be to develop and test protocols based on research participants whose facial features have been documented using the most accurate, up-to-date technology. Traditionally, data on the structures and tissues of the face was collected on cadavers using needle-puncture methods. However, due to the postmortem changes that occur to the body after death, such data collected may not accurately represent the appearance of the living individuals’ facial features. Now that access to technology such as CT scanning (including cone-beam CT for the head), MRIs, and ultrasounds has expanded, and the tools themselves are more precise in capturing detail, researchers are re-evaluating and re-developing prediction protocols using more accurate 3D image documentation. Testing techniques on different populations has also helped to grow our understanding of what techniques are most reliable to apply for specific cases, leading to the development of new protocols and the modification of existing standards to optimally estimate facial features for various demographics.
Natural variation in human anatomy forms the premises for both facial approximation and facial examination. Though these two forms of forensic identification are applied for different types of cases, it is crucial for both that practitioners have a strong background and understanding in facial anatomy, and how it translates to the characteristics we perceive for recognition and identification. Whether we are developing procedures for examining and comparing faces, or methods and protocols for estimating the appearance and dimensions of facial features, subjecting both existing and novel techniques to rigorous testing and validation will ensure that we are conducting accurate and thorough analyses that can be relied upon by the many stakeholders in forensic investigations.