1 Introduction

Anthropometry is one of the part of physical anthropology. It consists of methods that allow to describe the human body using quantitative traits. Its purpose is to collect comparable data (1) to describe individuals on the background of the entire population and (2) to compare individual groups. Anthropometry is based on body measurements of chords, arcs, angles, etc. determined by anthropometric points. Measurements can be performed on head (cephalometry) and body (somatometry). They are the basis for calculating anthropometric indicators that allow to classify collected data in subgroups. Anthropometry is widely used in areas such as medicine, in particular in plastic and reconstructive surgery, but also for planning surgical procedures; in the clothing industry, industrial design and ergonomics [2]. Currently, anthropometry is employed to identify people (passports and evidence with a biometric photo) or suspects in forensics [1].

Information technologies (IT), used in biometrics for identification or authentication, are generally used for image recognition. In the case of anthropometric features, the location of the points that are the mapping of anatomical sites should be determined. Their anatomical position is masked with constantly moving covering muscles and tendons. So the use of IT consists in the automatic acquisition of results, setting points similar to ROI (region of interest) and evaluation by an expert. In this way facial images are processed (automation of the results’ collection). In this context, IT is a common tool for supporting, with limited use in the extraction process of features.

Classic anthropometric studies are very laborious and take a long time. They require makings of many single measurement which take a lot of time, the specialistic equipment, the place and first of all uninterrupted presence of measured individual and measuring staff. The collection of substantial comparative material requires the involvement of many people for a long time. But it can be arduous, especially when children or people with disabilities will be examined [3]. Current technology could sometimes simplify this process, when photos, scans and 3D models for measurements are used.

Image recognition in biometrics is often used to identify or authenticate a person based on a planar mapping, formerly analog, now in the form of a finite state of the number of points. Examples of facial biometry methods using photographs are presented in [7]. The use of face identification techniques or its elements [11] makes sense rather semantic, than morphological. In the latter case, when ordered values of anthropological parameters are the basis for recognition, the homogeneity of the conditions of image acquisition plays a fundamental role. In this approach, the magnitude of errors symptomatic for quantitative measurements is a fundamental limitation of the method’s reliability. A small angular rotation causes significant changes in values describing the essential features. The indication of reference points is a particularly difficult issue; high levels of automation are expected because biometry should run in real time in its utilitarian sense. The expert’s help is undoubtedly priceless but time-consuming. Thus, the next problem is to validate the effectiveness of location algorithms for selected base points to determine the value of anthropometric parameters. Essentially, human plays this role; it is often expected that the task of validation will be carried out by a team of experts, assuming that the participation of independent reviewers will average and minimize the errors.

The aim of the study was to compare anthropometric indices obtained from measurements made by single measurer and team of measurers.

2 Materials and Methods

The material for analysis (part of LoVoiS – Longitudinal Voice Study) was measurements’ data achieving from two biometric photos: a boy’s aged about 15 years and girl’s at the age of 12. Both photos were taken using a Sony Alfa 350 camera with a Sony DT 18–70 mm f/3.5–5.6 lens using a Sony HVL-f42am lamp. The photographed individual had sat at a distance of 2 m from the lens, the lens axis was set to the tip of the subject’s nose, the camera’s vertical position was calibrated individually. The pictures were taken in the same resolution with the focal length set at 70 mm, with the same shutter opening time. The subject sat in an upright position with the head set in a horizontal Frankfurt plane (the axis passing through the lower edge of the eye socket and the upper edge of the ear tragus, parallel to the ground) and maintained a neutral facial expression. The pictures were taken in the same room, using a uniform white background behind the photographed person.

The photos were elaborated in ImageJ 1.52a program.Footnote 1 First, the picture was leveled to compensate for lateral head bending. For this purpose, the axis between the pupil centers was determined, its angle was examined relative to the bottom edge of the image and rotated to reduce the angle to 0 degrees. Next, thirty students (referred as measurers’ team) placed twice 26 anthropometric points in both photographs, next averaged (Table 1). These people were well trained in performing face measurements (during the half-year course of anthropometry they had marked anthropometric points on skulls, real heads, and facial images). Then one person (KG; referred as single measurer) 30 times repeated the selection of points. In order to avoid bias when measuring the same face, individual marks were made at several days intervals. The points marked in the picture are expressed by the positioning of the pixel in the XY coordinate system.

Table 1. Anthropometric points marked in the picture
Full size table

To calculate the distance between particular points, the Euclidean distance formula was used:

$$\begin{aligned} d=\sqrt{\left( \left( x_1-x_2 \right) ^2 +\left( y_1-y_2 \right) ^2 \right) } \end{aligned}$$
(1)

The following measurements were calculated and expressed in pixels: bizygomatic breadth (zy-zy), biogonial breadth (go-go), nose breadth (al-al), lip length (ch-ch), biocular diameter (ex-ex), interocular diameter (en-en), interpupillary distance (pp-pp), physiognomic face height (tr-gn), total face height (n-gn), top-facial physiognomic height (tr-sto), upper facial morphological height (n-sto), lower face height (sto-gn), subnasal height (sn-gn), nose height (n-sn), upper lip height (ls-sto), lower lip height (sto-li) and lip height (ls-li). Next such received measurements were used to calculate 10 indices of face proportions [8]:

  • the upper facial index ([n-sto]/[zy-zy] * 100),

  • total face index ([n-gn]/[zy-zy] * 100),

  • physiognomical face index ([tr-gn]/[zy-zy] * 100),

  • interocular index ([en-en]/[ex-ex] * 100),

  • nasal index ([al-al]/[n -sn] * 100),

  • naso-facial ([n-sn]/n-gn] * 100),

  • transverse naso-facial index ([n-sn]/[zes -z]] * 100),

  • lip index ( [ls-li]/[ch-ch] * 100),

  • mandibular-zygomatic index ([go-go]/[zy-zy] * 100),

  • mandibular-facial index ([go-go]/[n-gn] * 100).

Fig. 1.
figure 1

[source: M. Majchrowska with author’s modification]

Anthropometric points used in the measurements marked in the ImageJ program

Full size image

Before statistical analysis, all extreme values were removed and the distribution of individual data sets was examined. Depending on the distribution and size of the sample, differences between mean values of indices obtained from measurements made by single measurer and team of measurers were assessed, using tests of Kolmogorow-Smirnow, Welch, U Mann-Whitney and t-student.

3 Analysis and Conclusions

The comparison between mean values of anthropometric indices obtained from measurements made by single measurer and team of measurers had showed significant differences for almost all indices except the interocular index in boy’s photo (Table 2) and four indices in girl’s photo: total face, naso-facial, mandicular-zygomatic and mandibular-facial (Table 3).

Table 2. The comparison between mean values of anthropometric indices obtained from measurements made by single measurer and team of measurers; boy’s photo
Full size table
Table 3. The comparison between mean values of anthropometric indices obtained from measurements made by single measurer and team of measurers; girl’s photo
Full size table

In Figs. 2 (boy’s photo) and 3 (girl’s photo), the calculated values of standard deviations for each anthropometric index obtained by two analysed methods are compared. For boy’s photo the largest differences between the values of SD were found in the nasal index (ni), the lip index (li) and mandibular-facial index (mfi). SD for the nasal index was 1.56 and 3.65 for single measurer and for measurers’ team, respectively. This gap may be due to the difficulty of defining the exact position of the point nasion in the image. There is no clear marking edges that allow unambiguous definition of the point, and therefore the variations in position on the Y-axis may vary significantly. The analysed difference between both SD, for the lip index (li) is 1.82, what is surprisingly large. The labial edge is clearly marked and the location of the point’s border should not cause problems. It could be supposed that the reason for this is shading the area around the boy’s mouth, which may mislead the determining exact point’s position made by various people. The smallest differences between SDs occurred for the interocular index (ioi), which indicates a fairly high ease and repetitiveness of spotting in the eyes’ corners.

In the girl’s photo, the largest differences between SDs were also observed for the nasal index (ni), while the smallest differences occurred in the zygomatic-mandibular index (zmi).

Fig. 2.
figure 2

Comparison of the values of standard deviations for each anthropometric index obtained by two analysed methods- the boy’s photo (for abbreviations see Table 2

Full size image

The distributions of selected indices based the markings of single measurer and the measurers’ team were compared for the physiognomical face index (pfi) and mandibular-zygomatic index (mzi), and presented in Figs. 4 and 5, respectively. The mean values for the pfi were comparable, although SDs were different, with higher SD values for measurers’ team (Fig. 4). It proves the high accuracy made by both, single and team’s measurers. Whereas repeatability of the spottings made by single experts could be assessed as better than made by a group of experts. In turn, large discrepancies appeared when indices based on the go-go measurement were compared (f.ex. mandibular-zygomatic index, mzi). For boy’s photo there were visible differences between SD’s values and mean values obtained by two procedures. For girl’s photo differences between SD of measurements made by single measurer and measurers’ team were not large, the mean values of mandibular-zygomatic index (mzi) distinctly differ (Fig. 5) which points to low compliance and high repeatability. The anthropometric point gonion is difficult to determine when photos are analysed. The angle of the mandible (where this point is placed) is poorly visible in the en face photo, especially for women. It is the easiest to mark these paired points when heavy masculined faces have clearly defined mandible (i.e. male faces). It is possible that the measurers’ team more often assumed that in the photo, the gonion point is located in the place where neck’s and mandible’s lines are crossed. In turn, the single measurer more often placed the gonion point slightly above, probably in the most accurate way to project the anatomical gonion location.

Fig. 3.
figure 3

Comparison of the values of standard deviations for each anthropometric index obtained by two analysed methods- the girl’s photo (for abbreviations see Table 3)

Full size image

It can be concluded that single measurer adopts his/her own chart of the points and repeat their positioning, especially in the case of points difficult to define. It should be kept in mind especially when series of photos of the same individual will be analysed. Also creating databases of measurements performed by several people can lead to artificial differences in the analyzes and comparisons.

Fig. 4.
figure 4

Physiognomical face index values for single person measurements and team’s measurments for photos of a boy and a girl

Full size image
Fig. 5.
figure 5

Mandibular-zygomatic index values for single person measurments and team’s measurments for photos of a boy and a girl

Full size image

4 Summary

Nowadays biometric techniques constantly expand and improve in order to facilitate and accelerate work, research or design. Their significance constantly increases along with the creation of security of goods and the development of individual identification systems [9].

An appropriately calibrated image, with good-quality can be a research material that reduces the active participation of the subject and allows the elimination of errors created during traditional measurements [3]. Also permanent access to the photo permits to return to the material to improve, supplement or expand the scope of research. In the biometric face photo, points that are more or less equivalent to real anthropometric points can be successfully applied. It allows to perform correct calculation of indices and to assess real face proportions. Using photographs or cut frames from films, it is possible to examine the degree of face symmetry, sexual dimorphism and age-related changes [4,5,6, 10, 12]. An important aspect of this issue is the elimination of measurement error. So the repetitiveness of research techniques as well as the accuracy that single measurer is able to provide, is essential for repeatability of analyzes.

In the present case, the paradox of the higher reliability of a single expert over the team of experts has its source in the established rules of ordering points on the plane in terms of indicating anthropometric points. A photograph (analogue, but also digital) is characterized by an elusive border of particular fractions. The dark point may be darker than the nearby or lighter in specific lighting conditions, e.g. the protruding cheekbone is able to induce shading of the upper cheek in the case of the dominance of the upper lighting. An expert who constantly uses his/her own criteria at most commits a mistake with the character of a linear displacement error, relatively easy to correct, unless of course there are methods to correct the expert’s findings. The research will be carried out on the effectiveness of automatic search algorithms for anthropometric points.