Abstract: In this paper we discuss the ability to identify persons by ear from surveillance videotapes. The identification is through side-by-side comparison of surveillance videotapes. We present the results of a test, which was constructed to investigate whether the participants could individualize suspects by ear in a closed set situation. In general the possibility to identify a person by smaller body parts, in particular his/her ear, from surveillance videotape might become a useful tool as the availability of images from surveillance cameras’ is currently rapidly increasing. It is shown that the quality of the video images determines to a large extent the ability to identify a person in this test. The motivation for the test was identification of a person by ear from surveillance videotape in a case involving several gas-station robberies in Utrecht, The Netherlands.

Introduction

In The Netherlands as well as in many other countries the use of surveillance cameras has grown explosively. As a consequence an increasing number of crimes are recorded on video by surveillance cameras. Therefore, the question of identification of the person(s) on the tape is put to the forensic expert more and more frequently. However this can be a formidable task, especially when the recordings are of bad quality and large parts of the body are covered with clothes and or masks.

Sometimes an expert is asked whether he can make an identification based on a visible body part, for example, part of skin, hands, arms etc. Recently, a similar question was posed to the experts at our laboratory. It concerned several robberies of gas stations in Utrecht, The Netherlands. Video camera surveillance film was obtained from the scene of the crime. The offender was wearing a baseball cap, shawl and a cloth hanging from the cap to his face thereby covering his face. However, the ears of the offender were visible and the question was raised whether on basis of this material a certain suspect could be identified as being the offender.

For a possible identification two questions have to be answered. First, are ears unique or rare enough to come to enable individualization? Second, are the recordings of a quality good enough to distinguish sufficient details for individualization?

In this paper we discuss both questions. We conclude that the scientific answer to the first question is at this moment inconclusive, although there is at the moment some evidence in favor of uniqueness. However, this only holds if one can observe the ears in enough detail.

To investigate the second question we performed a small test motivated by the above case wherein we ask forensic experts, from other forensic disciplines but with a thorough knowledge of the individualization principles and processes, to individualize persons by ear from surveillance camera video film in a closed set situation. The closed set assumption is imposed to circumvent the uniqueness question, i.e. the ears of the individuals taking part in the test are distinguishable.

The results show that in this test the experts were able to determine whether they had sufficient information to make an individualization or exclusion. The results further indicate the importance of good quality surveillance equipment.

The paper is organized as follows, we first discuss some literature on ears and uniqueness, followed by a discussion of some issues regarding individualization and on the role of the surveillance camera system. Finally we describe our test and its outcomes and finishing the paper we some conclusions.

The question of uniqueness and classification of ears already has some history, although not always in a forensic context. We state some results, not exhaustive, from previous studies on ear biometrics and individualization. The most prominent is by Iannarelli (1989), he considered over 10.000 ears and found no indistinguishable ears. Imhofer (1906) already found that in set of 500 ears he only needed 4 characteristics to uniquely distinguish them. Some other references to ear identification, although some more connected to either ear prints or research to in heredity factors: Hirschi (1970), Hammer und Neubert (1989), Hunger and Hammer (1987), Oepen (1976), Rother (1976), Van der Lugt (1998).

Most of the above literature indicates that the variability between ears is that large that it might be possible that ears are unique, and moreover possibly uniquely distinguishable on a limited number of features or characteristics.

During the past decades in several countries, like The Netherlands, United Kingdom, West-Germany, Austria and the United States, already a substantial number of cases involving evidence based on ear prints and some isolated cases of video images of ears have appeared in court and in some cases accepted as evidence. We consider the aforementioned case, which our laboratory investigated, as one of these isolated cases. However, a formal validation of use of ear identification, being either ear prints or images of ears, has not taken place yet, to our knowledge. 

As identification (individualization) relies on uniqueness, the first hypothesis, which has to be tested, is whether ears are unique. As is well known such a hypothesis is very hard, if not impossible, to test formally and therefore the way forward is the same as in other biometrics identification issues such as for instance, finger- printing, facial recognition and DNA. The first step is the recognition that nature creates things and shapes with large between individual variation and that it might be the case that there is so much variation possible that an ear can be individualized. The second step is to collect as many observations as possible and check whether ears can be distinguished at a certain level of measured or observed accuracy. If it is the case that none of the observed ears can be classified as indistinguishable this is regarded as evidence in favor of the hypothesis that ears are unique. Final proof that the hypothesis is true however will never be obtained. Therefore as in many other biometrics identification problems uniqueness of ears is condemned to be an unproved working hypothesis.

The discussion on uniqueness is difficult and will probably go on for some time. We make the point that it is valuable that we already develop ideas and techniques that might be used once the uniqueness of ears has been established satisfactorily by scientific or legal standards. If it would turn out that it is not possible to develop good and well-validated procedures for individualization, the discussion on uniqueness would be left unimportant.

Besides the uniqueness or rarity of the ear, the quality of the video images and the associated possibility to distinguish different characteristics of the ear are of utmost importance, the images must enable one to individualize the ear, which means, by Tuthill (1994):

We will consider some of the factors that influence the quality of the images in the next paragraph but make here some comments on the process of individualization or recognition based on the available images.

As we were only interested in whether the experts were able to correctly identify or exclude persons as being the same in two different video clips based on ear characteristics, we did not ask the experts on how they came to their conclusion. This means that the reasoning on how they came to individualization was not verified, which of course in practice is of utmost importance. However we leave this aspect of the individualization process for further research.

We further note that according to the definition of individualization above, individualization has to be performed based on precise defined characteristics or points (of which also the probability of random occurrence is (approximately) known). In photographic and video film comparisons, like the case above, it is not always possible to define exact characteristics, or only with a great loss of information. In test above we were actually interested in the ability of the test respondents to make the decision of individualization making use of all available information, definable or not.

We note a way around the dependence on precise defined points is often the expert’s opinion. However, in such cases he/she will often not be able to give information on the quality of his/her opinion although it might be entirely correct. To get around this problem one might put the problem, whether or not put in the form of a (double) blind test, to a group of independent proven experts, collect their opinions and if they agree present their opinion as evidence. This way one is able to get some idea about the relative value of the evidence.

In general the ability to individualize increases if the number of observable details in the images of the surveillance camera increases. Firstly, this should be an incentive to install good quality cameras and recording equipment. Secondly, it points out that it is worth thinking carefully about the positioning of the cameras and the availability of light. In practice however, numerous systems have been installed not optimally with respect to the positioning of the camera and available light. Further it frequently occurs that the recording equipment is badly maintained and lightning changed after installation of the system. From a forensic point of view it might be wise for the users of surveillance cameras to prevent the above problems as much as possible as it makes a successful individualization more likely.

For the forensic expert however, these factors are given. The quality of the images is therefore influenced by a large number of uncontrollable factors. Some of the factors are associated with the surveillance system, such as the position of the camera, distance camera – offender, quality of the tape, type of camera, etc. Extremely important is the available light, specially as the ear is relatively small and rather curved its the light falling on the ear which determines whether there is enough contrast to distinguish features from false shadows. Other uncontrollable factors are associated with the behavior of the offender such as his/her clothing, hair, disguise, movements with respect to the lens etc. All of these factors add possible problems to the process of individualization.

In the test described below we consider video images from two different gas stations with installed surveillance equipment. By doing so we hope to be able to evaluate the effect of the uncontrollable differences between the stations and the resulting quality of the images.

Description of the test

To investigate whether it is possible to identify a person by his/her ear and to get an idea about possible differences between experts and what factors play a role in the individualization process a small experiment was carried out.

Surveillance camera video clips were recorded at two of the gas stations that were robbed. The standard surveillance equipment present, Times Laps surveillance cameras, was used to make the recordings using on a VHS-video recorder. From the Utrecht police department 22 persons, all male, volunteered to cooperate. They were asked to make a set of prescribed moves. At each station a group of the men were put in front of the cameras wearing the same type of overall. This was done in three badges; the first badge was considered as perpetrators wearing a mask, the second badge as masked suspects and the third badge as unmasked suspects.

The setup was motivated by several aspects, first by varying the stations we were able to capture different camera positions, lightning, position and movements of the perpetrator and other uncontrollable factors as mentioned above. Second, the choice to have unmasked suspects too was that in the actual case the identification had to be done from comparing a video with a masked robber with an unmasked suspect. Third, the men were only available for one evening, this meant a restriction on the total number of video clips that could be made.

Before presenting the video clips to the participants performing the individualization the clips where digitalized and put on a computer. To avoid being influenced by other characteristics of the person, height, size, way of moving etc, a mask was put over the video with whole showing the ear and as few as possible further aspects. Finally, the whole was centered in the screen.

As formal ear-experts are not available and the main goal of the test was to investigate whether the quality of the images was sufficient for individualization we asked 6 forensic experts to do the test. They all are well trained in the individualization process in their own field of expertise.

The respondents participating in the test were presented 40 sets of two video film clips and asked two questions. First, do you think there is enough information in these video films to make an individualization or exclusion? Second, they were asked whether the persons in the videos where the same or not. If they answered the first question positive the answer to the second question were interpreted as "hard" while if they answered the first question negative their answer to the second question were interpreted as an informed guess. The motivation for this line of questioning was twofold. We first of all wanted to know how much information there actually was in the films, whether this information was picked up conscious or unconscious. Second, we were interested whether the respondents were able to determine whether they had information enough for individualization.

Before the respondents started the test they were given for samples to get acquainted with video film ear comparison. They were also told that ears of different people were different form each other, i.e. we introduced the closed set assumption.

From the available films, 17 from station 1 and 22 from station 2, 40 sets of two video film clips were selected. In each couple both clips came from the same station. The randomization was performed by first putting the available combinations in a list, maximizing the number of couples with the same person in each of the clips, then selecting randomly without replacement from this list. We note that some of the persons featured in different clips. The distribution of the selected combinations is given in Table 1.

The same test was given to each of the 6 respondents. The results are presented in the Table 2-4. When interpreting the results one should keep in mind that the video films from station 2 were clearly of less quality the those of station 1.

The results of Table 2 are not very promising, as there appear to be a significant percentage of false identifications. Although, formal Chi-square tests, both have p-values <0.01, show that the choice is clearly better than random.

Things brighten up when we consider Tables 3 and 4. The respondents decided that in 39 out of the 240 comparisons they had sufficient information for individualization. Once the respondents had decided they had sufficient information, they made no false identification on either station. Further, the respondents make 10.2% wrong decisions if they think to have sufficient information while they make 25.8% of wrong decision in cases where they indicate not to have sufficient information.

From Table 4, in case of respondents think to have insufficient information, Chi-square tests indicate that the hypothesis of random choice is still rejected strongly, p-values <0.01. However there are relatively many false identifications, indicating a diagnostic value of an individualization of 2.1 for station 1 and 2.9 for station 2 respectively.

The distribution of the wrong decisions is given in Figure 1.

There clearly is an indication of differences between the respondents. Respondent 1 made half the errors the respondent 6 made. Further research should investigate of in larger experiments the same holds and what causes this difference.

To conclude the experiment clearly shows that better images lead to more and more correct individualizations. Clearly an argument to ask for better surveillance equipment. Further we find that, although the experiment is of limited size, that the respondents were able to determine whether there was sufficient information for individualization. Finally, in this closed set situation there is an indication that once respondents indicate there is sufficient information a positive individualization has a strong evidential value.

Conclusions

In this paper we discussed the possible use of identification by ear from surveillance camera film. First, the uniqueness of ears is discussed, it is argued that the there is some evidence in the literature which is in favor of the uniqueness of ears if enough detail is taken into account. However, it is pointed out that more evidence needs to be gathered to make it a sound acceptable working hypothesis. Further, we investigated with a small test whether in a closed set situation respondents can individualize persons by ear from surveillance camera film and whether they can decide whether they have sufficient information to come to an individualization or exclusion. Both questions can be answered positively for this test, although the first less strongly than the second question.

We further conclude that the forensic community has to make clear to users of camera surveillance equipment that better equipment together with proper installation and handling will payoff by a larger probability of actually being able to identify offenders.

Finally we think that forensic science could benefit from the introduction of double blind tests in future cases as these give an immediate indication on the reliability of the procedure and/or the expert. Especially, when considering types of evidence which is still based on less formalized models and procedures.

Acknowledgements

In this research numerous people have participated therefore we wish to thank. First of all the Utrecht Police Department organizing the shooting of the film clips for the test; Adee Schoonen for advise on the design of the test; Jurrien Bijholds for his expertise in digital imaging; and Marjan Sjerps for many helpful discussions. We especially thank the NFI people that cooperated in the test.

Return to Master Index