Human Face Detection - How Well Can You Recognize Faces?
Face detection in humans is a complex process which we have come to depend on. Recognition is how our brain creates and compares descriptions of objects we can see in front of us with descriptions of objects that we have seen previously.
In psychology research, face detection is abundant with theories on the mechanisms which drive this ability. Furthermore, those who are unable to recognize faces at all, termed 'prosopagnosia', provide additional information regarding the processes which may be at work.
Recognition starts with how we recognize objects in our everyday world. This involves a number of clear stages involving perception, categorization and naming, as defined by Humphreys and Bruce (1989).
The object naming stage allows us to recognize objects in different ways:
Between-category distinctions: where we name the category the object is in i.e. fruit or furniture.
Within-category distinctions: where we identify the object within that category i.e. for faces, we don’t say ‘faces’ we work out whose face it is.
Much research has centred on whether faces are recognized by the same processes used to recognize objects. The answer has not yet been found but the difference between category distinctions is why face detection is normally studied as separate topic to object recognition.
In face detection, there are unique issues to be considered, namely:
- A face can move, which in turn changes its appearance
- such movement can express social or emotional cues
- faces can dramatically change over time, such as through hair cuts and aging
There are also many different types of face detection which sets it apart from other recognition processes, for example recognising familiar and unfamiliar faces.
Facial Expressions and Emotions
Generally, we are able to recognize the face we are looking at and the emotion that it is portraying. Faces are very important in conveying emotional state; we are able to judge emotion very accurately from a face and we are very sensitive to eye movements in those around us.
Young et al (1993) claimed we have specific processes for recognizing emotions but these processes are not involved in recognizing identity.
We are able to tell if a person is angry or happy even if we don’t recognize them, and we need to be able to recognize people in these different emotional states with different facial expressions.
Case Study: Cashiers and Shoppers
Kemp et al (1997) studied how well cashiers match shoppers to credit cards bearing their photographs.
They found that cashiers often accept cards with photos which only bore a resemblance to a shopper and even accepted cards with no resemblance but they were the same gender and ethnic background.
Faces can be classed at different levels. We can:
- decide that stimulus is a face as opposed to an object
- decide if face is male or female
- decide on ethic origin and other characteristics
- decide if face is familiar or unfamiliar
Such within-category judgement sets face recognition apart from object recognition and it is considered more visually demanding because such minimal differences can be present between faces.
Face recognition is a similar matching process to object recognition but there is the need to access relevant semantic information and a person’s name.
Diary Studies of Face Detection
Young et al (1985) conducted a diary study where 22 participants were asked to note mistakes they made in recognizing people over an eight week period. The categories these mistakes fell into were:
- Person misidentified: someone unfamiliar misidentified as someone familiar
- Person unrecognised: someone familiar thought to be someone unfamiliar
Both of these could occur due to poor viewing conditions, for example it is dark or if you do not know person very well.
Errors in Face Detection
- Person seemed familiar only: recognized as familiar but no other information about them is remembered immediately
- Difficulty in retrieving full details of person: only some semantic information retrieved but not specifics such as their name
These errors tend to occur when a familiar person is seen outside the context they are normally seen.
The pattern of these errors suggest that despite the fact we may retrieve previously learnt semantic information about a person without recalling their name – it will never happen other way round – we will never recall a name without recalling relevant semantic information about the person. However a key point is, before any of this can happen, we must detect that the face is familiar to us.
Case Study: School Teachers and Students
In 1984 Bahrick studied school teachers recognition of former students who they had taught over ten weeks, at between 3 to 5 times a week.
The level of face recognition for those they had taught recently was high, at 69%. This dropped as the number of intervening years increased. After 8 years only 26% of former students were correctly recognized.
Lab studies lend support to the notion that different types of information is sequentially accessed.
Hay et al (1991) showed participants 190 famous and unfamiliar faces and asked them to decide if each face was familiar and to state the person’s occupation and their name.
Participants did not retrieve a name without their occupation which supports the idea that semantic identity information is retrieved before a name.
Face Recognition System
Such findings are consistent with the notion that face detection involves a sequence of processes using different types of information. Young et al (1985) refined a cognitive theoretical framework where recognizing a person involves sequences.
On meeting people, we encode their faces which may activate face recognition units (FRUs) which contain stored information about faces we are familiar with. If there is a match then recognition units are activated and allow access to semantic information about person’s identity stored in person identity nodes (PINs). Only once a PIN is activated can a name be generated.
IAC Face Recognition Model
Bruce and Young (1986) proposed a similar model where face recognition occurs in clear sequential stages.
In 1990, Burton and Bruce proposed the Interactive Activation and Competition (IAC) Model which was very much an extension of Bruce and Young’s work. This model suggests that the sequential stages involved are interconnected in an interactive network, hence the term Interactive Activation and Competition. They included semantic information units (SIUs) in the model and suggested FRUs, PINs and SIUs all result in a lexical output representing either words or a name regarding the person in question.
Burton and Bruce (1990) IAC Model of Face Recognition
The pools are connected by the input systems (FRUs) which join to a common set of person identity nodes (PINs) and these are linked to units containing semantic information (SIUs).
All this information combined works together in an inhibitory and excitatory manner throughout the network until the recognition process is complete. This model explains Young’s diary study results and the use of additional semantic information in the face recognition process.
Face Blindness - 'Prosopagnosia'
Prosopagnosia is the inability to recognise faces whilst maintaining the ability to recognise other objects. Also known as ‘face blindness’, pure prosopagnosia is very rare and there are normally other deficits present.
Key findings from investigation of prosopagnosia:
- Identification of expression appears to be independent from face identification
- Face recognition and awareness of it may also be independent of one another
In many cases, the ability to recognise facial expressions may be unaffected.
Examples of Prosopagnosia Cases
Bauer (1984) studied patients with prosopagnosia and used the skin conductance response (SCR) to monitor changes in automatic nervous system activity when carrying out face recognition tasks. Changes in SCR during such tasks would signal an emotional reaction to stimuli regardless of conscious processing.
One patient, LF, was shown a face and read a list of 5 names while their SCR was measured. When LF was asked to pick the correct name for the faces he was looking at, he was unable to recognise familiar people from their faces alone. However, LF showed a greater SCR when the correct name was read aloud compared to incorrect names. This suggests that LF was emotionally responding but was not conscious of this response enough to recognise the people in the pictures in terms of their names. This has been termed ‘covert recognition’.
Case Study: Bilateral Brain Injuries
Young et al (1993) conducted a study of ex-serviceman with bilateral brain injuries.
They found subjects with a right hemisphere lesion were selectively impaired in the identification of familiar faces. One subject with the same damage had problems only with matching unfamiliar faces and a number of subjects with left hemisphere damage were found only to be impaired on facial expression tasks.
It is thought that provoked overt recognition can occur under experimental conditions.
Sergent and Poncet (1990) studied a patient 'PV'. When PV was shown 8 faces of famous people, she was unable to identify them.
However, when told they all had the same occupation and she looked at the faces again, she was able to identify they were all politicians and name 7 of them.
IAC and Covert Recognition
Covert recognition fits with the IAC model in that it could be an example of a weakening in connections between FRUs and PINs. For example, excitation of a corresponding PIN is not raised above the threshold for a face to be recognized.
By informing the patient that the faces are all related by occupation is equivalent to strengthening the PIN to SIU connections. Once strengthened, activation is passed back from shared SIUs to relevant PINs which then activate the threshold and faces are fully recognised.
Another interesting finding with face detection research is the 'inversion effect'. This is where inverting or turning upside down visual stimuli impairs our ability to recognise faces compared with ability to recognise objects.
Diamond and Carey (1986) claimed the inversion effect is due to our perceptual mechanisms becoming used to seeing this type of stimuli in a visual upright orientation, therefore this ‘tuning’ is lost when we see a face inverted.
The Complexity of Face Recognition
The case study of PV is useful as it highlights how semantic information of occupation helped the patient to access name information. In the IAC model this would be explained by this information flowing through the network adding information, for example eliminating some possibilities if they did not fit that occupation and highlight others that did. Links are therefore increased which leads to the final accurate face recognition.
Evidence from those who have prosopagnosia provides interesting additional information about how our face detection system may operate, in what is clearly a complex series of mechanisms which come together to aid our ability to recognize the people around us.
- Memory Psychology - The Role of Cognition and Emotion
The study of memory in psychology is a rapidly advancing area of research. The interconnection of cognition, emotion and memory has been particularly insightful in moving this area forward.
- Perception Psychology - How We Understand Our World
Perception in psychology enables us to make sense of the world around us. We view our surroundings and this information is translated into meaning within the brain.
- Cognitive Neuropsychology - Discoveries of Broca and Wernicke
Neuropsychology is concerned with the brain and its interactions with the cognitive functions of the mind. Broca and Wernicke both made key discoveries vital to the development of this discipline.
- Bahrick, H. P. (1984) "Memory for people" Everyday memory, actions and absent-mindedness, 19-34.
- Bauer, R. M. (1984) "Autonomic recognition of names and faces in prosopagnosia: A neuropsychological application of the guilty knowledge test" Neuropsychologia, 22(4), 457-469.
- Bruce, V., & Young, A. (1986). "Understanding face recognition" British journal of psychology, 77(3), 305-327.
- Burton, A. M., Bruce, V., & Johnston, R. A. (1990) "Understanding face recognition with an interactive activation model" British Journal of Psychology,81(3), 361-380.
- Diamond, R., & Carey, S. (1986) "Why faces are and are not special: an effect of expertise" Journal of Experimental Psychology: General, 115(2), 107.
- Hay, D. C., Young, A. W., & Ellis, A. W. (1991) "Routes through the face recognition system" The Quarterly Journal of Experimental Psychology, 43(4), 761-791.
- Humphreys, G. W., & Bruce, V. (1989). Visual cognition.
- Sergent, J., & Poncet, M. (1990) "From covert to overt recognition of faces in a prosopagnosic patient" Brain, 113(4), 989-1004.
- Young, A. W., Hay, D. C. & Ellis, A. W. (1985) "The faces that launched a thousand slips: Everyday difficulties and errors in recognizing people" British Journal of Psychology, 76, 495-523.
- Young, A. W., Newcombe, F., DeHaan, E., Small, M. & Hay, D. C. (1993) "Face perception after brain injury: selective impairments affecting identity and expression" Brain,116, 941-959.
© 2015 Fiona Guy