Bloodstain Pattern Analysis

A New Challenge for Computational Intelligence Community

Giovanni Acampora

, Autilia Vitiello

, Ciro Di Nunzio

, Maurizio Saliva

and Luciano Garofano

5,6

School of Science and Technology, Nottingham Trent University, Clifton Lane, Nottingham, U.K.

Department of Computer Science, University of Salerno, Via San Giovanni Paolo II, Fisciano, Italy

University ”Magna Graecia” of Catanzaro, Catanzaro, Italy

Azienda Sanitaria Locale ASL Napoli 3 Sud, Torre del Greco, Italy

Arma dei Carabinieri, Italy

Italian Academy of Forensic Sciences, Reggio Emilia, Italy

Keywords:

Forensic Intelligence, Pattern Recognition, Image Processing, Fuzzy Reasoning.

Abstract:

Bloodstain pattern analysis (BPA) is a forensic discipline that plays a key role in tracing events which caused

a bloodshed at a crime scene. Indeed, BPA supports worldwide investigation agencies (US FBI, Italian Cara-

binieri and so on) in interpreting the morphology and distribution of bloodspots at a crime scene in order to

enable a potentially complete reconstruction of the dynamics of the act of violence with a consequent iden-

tiﬁcation of potential suspects for that crime. However, in spite of its importance, this forensic discipline is

still based on completely manual approaches, making the analysis of a crime scene long, tedious and poten-

tially imperfect. This position paper is aimed at proving that computational intelligence methodologies can

be efﬁciently integrated with image processing techniques to support forensic investigators in increasing their

performance in examining bloodstains, both in terms of time and accuracy of analysis. A preliminary study

involving the application of fuzzy clustering has been carried out in order to validate our opinion and stimu-

late computational intelligence community to face this new challenge towards a formal deﬁnition of Forensic

Intelligence.

1 INTRODUCTION

Bloody crimes among strangers, acquaintances or

family members are becoming more and more fre-

quent in our society, resulting in a growing sense of

fear in the population with related sociological and

relational issues. In order to address this increasing

wave of violence, worldwide investigation agencies

are making strong efforts to improve their abilities

in analysing crime scenes through sophisticated sci-

entiﬁc methods aimed at efﬁciently solving complex

cases and acting as a deterrent to violent crimes. In

this novel investigation scenario, Bloodstain Pattern

Analysis (BPA) is assuming a crucial role thanks to its

potential skills in identifying precious clues useful for

the complete reconstruction of the dynamics of acts

of violence. Precisely, BPA is a forensic investiga-

tion discipline that deals with the analysis of morphol-

ogy and distribution of bloodstains at crime scene.

Its principal aim is to shed light on various forensic

matters including reconstruction of events, differen-

tial diagnosis of homicide/suicide/accident and iden-

tifying areas with high likelihood of offender move-

ments for taking DNA samples. The ﬁrst systematic

study of bloodstains was published in 1895 by Eduard

Piotrowski from the University of Krakow. In this

study, entitled “On the formation, form, direction and

spreading of blood stains resulting from blunt trauma

at the head” at the University of Vienna, Piotrowski

covered the corner of a room with sheets of white

paper and observed and documented the bloodstains

that resulted from beating rabbits to death (Brodbeck,

2012). Since that time, BPA has become an estab-

lished analytical technique in forensic investigations

and the International Association of Bloodstain Pat-

tern Analysts (IABPA) was founded to support the

continuing development of the discipline. However,

in spite of its importance, BPA is still based on a

fully manual approach, making the analysis of a crime

scene long, tedious and potentially imperfect. Indeed,

211

Acampora G., Vitiello A., Di Nunzio C., Saliva M. and Garofano L..

Bloodstain Pattern Analysis - A New Challenge for Computational Intelligence Community.

DOI: 10.5220/0005155602110216

In Proceedings of the International Conference on Fuzzy Computation Theory and Applications (FCTA-2014), pages 211-216

ISBN: 978-989-758-053-6

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

by following the current BPA guidelines, investigators

use their experience to manually identify blood spat-

ter patterns and perform geometrical measures to lo-

cate the point of origin, i.e. the spatial location where

the identiﬁed bloodstains has been originated. The

collection of points of origin is then used by investi-

gators to try to determine the full dynamics of events

at the crime scene.

This position paper is aimed at proving that com-

putational intelligence methodologies can be efﬁ-

ciently integrated with image processing techniques

to support forensic investigators in increasing their

performance in applying BPA, both in terms of time

and accuracy of analysis. In particular, image pro-

cessing techniques can be used to capture pictures

from a crime scene, remove noise, register that pic-

tures and extract the collection of features that com-

putational intelligence methods can efﬁciently anal-

yse to make BPA faster and more precise than cur-

rent manual approaches. A preliminary study involv-

ing the application of fuzzy clustering for reproducing

the well-known string method has been carried out in

order to validate our opinion and stimulate computa-

tional intelligence community to face this new chal-

lenge towards a formal deﬁnition of Forensic Intelli-

gence.

2 BLOODSTAIN PATTERN

ANALYSIS

Blood is one of the most signiﬁcant and frequently en-

countered types of physical evidence associated with

a violent crime (James et al., 2005). Consequently,

forensic investigators use a formal methodology, the

BPA, to assesses bloodstains left at crime scenes by

using an approach based on visual pattern recognition

(Brodbeck, 2012). Thanks to this visual approach,

BPA investigators analyse the size, shape, and distri-

bution of bloodstains resulting from bloodshed events

in order to determine the types of activities and mech-

anisms that produced them. In particular, BPA may

provide several types of information to forensic in-

vestigators as, for example (James et al., 2005): 1)

areas of convergence and origin of the bloodstains,

2) type and direction of impact that produced blood-

stains or spatter, 3) mechanisms by which spatter

patterns were produced, 4) assistance with the un-

derstanding of how bloodstains were deposited onto

items of evidence, 5) possible position of victim, as-

sailant, or objects at the scene during bloodshed, 6)

possible movement and direction of victim, assailant,

or objects at the scene after bloodshed, support or

contradiction of statements given by accused and/or

witnesses, 7) additional criteria for estimation of post-

mortem interval. Moreover, BPA is used to shed light

on other forensic matters such as differential diagno-

sis of homicide/suicide/accident and identifying areas

with high likelihood of offender movements for tak-

ing DNA samples.

BPA activities are based on a bloodstain classiﬁ-

cation from S. James, P. Kish and P. Sutton (James

et al., 2005) which divides bloodstains into three cat-

egories, passive/gravity, spatter and altered based on

stain physical features of size, shape, location, con-

centration, and distribution (Brodbeck, 2012). In de-

tail, passive category describes bloodstain patterns

that are formed under the inﬂuence of gravity. This

group includes contact stains, which result from con-

tact between two surfaces, of which at least one has

blood on it. Contact stains often provide information

about sequences of movement. Flow patterns, pool-

ing/saturation and drip stains also belong to this cat-

egory. Spatter category includes spatters that result

from active events such as a shot, as well as spatters

that are caused by, for example, expiration or cast-off

from objects that are swung. Altered category con-

tains all further stain types, such as blood clots and

diluted blood that results from the addition of other

liquids.

All the BPA analysis depends on the fact that

blood is a complex non-Newtonian viscoelastic ﬂuid.

For this reason, a drop of blood tends to form into

a sphere rather than a teardrop shape when in ﬂight.

Ideally, once the sphere lands on a ﬂat surface, the

collision ﬂattens the liquid creating an elliptical or cir-

cular stain depending on the angle of impact. The an-

gle of impact is the angle at which a blood droplet im-

pacts a surface, measured with respect to a imaginary

line perpendicular to that surface. In particular, the

more acute the angle of impact, the greater the elon-

gation of the bloodstain as the width decreases and the

length increases (James et al., 2005) (see Fig. 1).

Figure 1: Elongation of bloodstains in terms of the angle of

impact

Starting from the aforementioned bloodstain clas-

siﬁcation and the physical features of blood, BPA an-

http://science.howstuffworks.com/bloodstain-pattern-

analysis3.htm

FCTA2014-InternationalConferenceonFuzzyComputationTheoryandApplications

212

alyst can determine the impact angle of blood on a

ﬂat surface by evaluating the shape of the blood spat-

ter stain through trigonometric principles. In detail,

the analyst has to locate each spatter and measure its

length L (major diameter) and width W (minor diam-

eter) using a scale, a ruler or calipers (see Fig. 2).

Then, he or she computes the angle of impact α by

using the following formula:

α = arcsin

(1)

BPA analysis focuses only on primary stains (the

stain obtained when the blood drop just touches a sur-

face), however, it is worth noting that the collision is

a complex interaction that produces also non-primary

spatter patterns caused by displacement, dispersion,

and retraction processes. Moreover, target surface ab-

sorbency, surface texture, and blood volume are im-

portant variable in blood pattern formation and BPA

analysts need to adapt their evaluations to these dif-

ferent factors.

Figure 2: Schematic image (Boonkhong et al., 2010) of the

bloodstain from a blood droplet with an impact angle α. In

the calculation of α, an analogy is made between line c and

b and the width W and length L of the stain.

The consolidated method to determine the point

from which the blood originated, named point of ori-

gin, is known as string method (James et al., 2005).

The name of this method comes from the way in

which forensic experts analyse the bloodstains (see

Fig. 3). In particular, once identiﬁed a spatter pat-

tern, the analyst attaches elastic strings at the tip

of the bloodstains’ ellipse and extends them back-

ward, or 180

◦

opposite of their individual directions

of travel. The two dimensional point, named point or

area of convergence, where the strings intersect rep-

resents the two-dimensional geographic location of

blood source. Determining the area or point of origin

combines the two-dimensional area of convergence

plus the angle of impact α discussed above for each of

the stains belonging to the pattern. In detail, analysts

pull elastic strings from the surface according to the

angle α. In this way, the angle of impact adds the third

dimension to the point of convergence determination,

creating a spatial representation of the location of the

blood source. This method gives forensic experts an

upper bound on the height at which the victim was

struck. However, repeating this process for up to hun-

dreds of blood stains takes substantial time and effort

(Shen et al., 2006). Moreover, it assumes the identi-

ﬁcation of a spatter pattern based on forensic experts’

knowledge.

Figure 3: Analyst performing the string method (James

et al., 2005).

3 COMPUTATIONAL

INTELLIGENCE FOR

BLOODSTAIN PATTERN

ANALYSIS

As highlighted in Section 2, BPA and, in particular

the string method, represents a crucial activity in each

investigation task related to a violent crime. However,

BPA activities could be affected by a set of strong

uncertainties due to different factors that can make

the overall forensic analysis not signiﬁcant. Indeed,

ﬁrstly, a crime scene is inherently imprecise due to

high interaction occurring among victim, aggressor

and surrounding environment. Moreover, the string

method is performed manually by means of impre-

cise tools and, as a consequence, each step related

to this activity adds more and more inaccuracies. In

our vision, all this imprecision make BPA an applica-

tion domain particularly suitable to be addressed by

computational intelligence techniques. Indeed, fuzzy

reasoning methods could be efﬁciently used to sup-

port pattern recognition activities in BPA in order to

analyse the shape of blood spots and compute conver-

gence points and areas, and points of origin of hits.

BloodstainPatternAnalysis-ANewChallengeforComputationalIntelligenceCommunity

213

At the same way, neural networks and other learn-

ing approaches could be used to automatically clas-

sify bloodstains by taking into account the heuristic

classiﬁcation of bloodstains perfumed by James et al.

2005. Finally, optimisation capabilities provided by

evolutionary algorithms could be useful to derive the

most suitable path followed by victim and aggressor

in a crime scene that is characterised by the presence

of an appropriatelyanalysed collection of bloodstains.

These are just few examples of application of compu-

tational intelligence techniques to BPA but, however,

in our opinion a very fruitful research path could start

in this scenario, making BPA as the mainstream ap-

plication of Forensic Intelligence area. Fig. 4 shows

our idea for a general template of a system architec-

ture where image processing and computational intel-

ligence techniques are integrated in forensic investi-

gation activities to implement intelligent BPA tools.

Intelligent BPA

Crime Scene

Observation and Analysis of

Evident Blood Traces

Highlighting (Luminol), Observation

and Analysis of Latent Blood Traces

Image Processing and Computer Vision

Computational Intelligence Methodologies

Documentation and Validation

Photos from Crime Scene

Bloodstain Features

Automatic BPA Report

Figure 4: Overview of the proposed generic architecture for

intelligent BPA.

In order to validate our vision, an embryonic in-

telligent tool for BPA is introduced. This tool uses

a computer vision algorithm for image rectiﬁcation

and the fuzzy C-Means clustering algorithm extended

with the silhouette approach for identifying differ-

ent bloodstain patterns present in a given image and,

for each pattern, computes the convergence area and

point of origin. In other words, the proposed tool

represents the ﬁrst attempt to make the BPA string

method completely automatic and unrelated to impre-

cise tools and tasks.

3.1 An Intelligent Tool for BPA

This section presents an embryonic tool designed to

perform the string method in an automatic, fast and

precise way. The proposed system performs a se-

quence of tasks summarized in Fig. 5. In particular,

Figure 5: Overview of tasks performed by the proposed sys-

tem.

the ﬁrst task consists in registering all photographs of

a crime scene to a common virtual image.

Indeed, in order perform an opportune analy-

sis, the bloodstain images should satisfy the follow-

ing constraint: to be photographed with the image

plane parallel to the surface where the bloodstains im-

pacted. Unfortunately, often the captured images at a

crime scene have not this feature. Therefore, the pro-

posed system performs a rectiﬁcation process through

image processing techniques, as described in (Shen

et al., 2006), in order to obtain to look as if the cam-

era had been looking down at the crime scene from

directly above. Fig. 6 presents an example of a pho-

tograph subject to the rectiﬁcation process.

(a) Original photo (b) Relative rectiﬁed photo

Figure 6: Example of the rectiﬁcation phase.

After the rectiﬁcation phase, the proposed system

executes a direct least squares ﬁtting method (Fitzgib-

bon et al., 1999) in order to identify blood spots in a

photograph. Fig. 7 shows a photograph where the

FCTA2014-InternationalConferenceonFuzzyComputationTheoryandApplications

214

Figure 7: Image where the dotted lines show the identiﬁed

blood spots.

dotted lines are the identiﬁed blood spots.

Successively, the proposed system computes the

major and minor diameters of each identiﬁed blood

spot. In detail, for the i

identiﬁed blood spot, the

proposed method computes two points p

= (x

, y

)

and p

= (x

, y

) corresponding to the farthest points

on the ellipse related to the bloodstain under analysis

and two points p

= (x

, y

) and p

= (x

, y

) cor-

responding to the nearest points. The distance be-

tween the points p

and p

represents the length of

the i

blood stain L

, whereas, the distance between

the points p

and p

represents the width W

. L

and

are used to compute the angle of impact of the i

bloodstain as described in section 2.

Then, the successive step performed by the pro-

posed approach is to build a dataset useful for the the

identiﬁcation of the number of blood spatter patterns

in a photograph. Firstly, the proposed system uses

the conventional formula to compute a straight line

through the farthest two points p

and p

for each

blood i:

y− y

− y

x− x

− x

Then, the system gathers a set of points F, where

each point represents the cross between two drawn

lines. These points, denoted as pairwise points of con-

vergence, represent the blood source for the involved

couple of bloodstains. Fig. 8 shows the building of

the collection of the pairwise points of convergence.

Once a photographhas been registered and a set of

pairwise points of convergencehas been collected, the

system exploits a fuzzy clustering procedure extended

with silhouette method (Rousseeuw, 1987) to iden-

tify the number of the blood spatter patterns present

in the photograph and the relative points of conver-

gence. These points of convergence together with the

previously computed values of angles of impact will

be used to compute the points of origin. In detail,

the proposed system computes the number of blood

Figure 8: Pairwise points of convergence computation.

spatter patterns as the number of obtained clusters

and the relative points of convergence as the center of

the correspondent cluster. The exploited fuzzy clus-

tering procedure is the well-known Fuzzy C-Means

(Bezdek, 1981) which allows one piece of data to be-

long to two or more clusters with a different member-

ship degree. This feature is useful in an environment

characterized by high uncertainty such as a crime

scene. The dataset in input of the Fuzzy C-Means

are the set of pairwise points of convergence. The

most opportune number of clusters in which dataset

must be divided is computed by using the silhouette

method. In detail, the proposed approach performs

the Fuzzy C-Means M times by using each time a dif-

ferent number of clusters k ∈ {1, 2, . . . , M}, where M

is the number of bloodstains in the photograph. This

upper bound for value k is chosen by reﬂecting that

there is at least a bloodstain for each spatter pattern,

and as a consequence, the number of spatter patterns

can not be greater than the number of bloodstains.

Then, the best number of clusters is chosen by apply-

ing the silhouette method, i.e., it is chosen the number

of clusters which allows to obtain the greatest value

of the overall average silhouette width (Rousseeuw,

1987). By considering our example, Fig. 9 shows

the values of the overall average silhouette widths for

each tested k = 2, . . . , 9. As shown, in our example,

the best number of clusters in which dataset should

be divided is 3. Fig. 10 shows the output of the fuzzy

clustering procedure for k = 3.

However, as highlighted in (Rousseeuw, 1987),

one should not merely accept a high overall average

silhouette width, but also look the output of cluster-

ing procedure in order to observe the presence of out-

lier values. Indeed, when a cluster contains only a

point of the input dataset, it is very probable that this

point is an outlier. Therefore, the proposed system

removes singleton clusters. Hence, in our example

characterised by one singleton cluster (see Fig. 10),

the resulting number of blood spatter patterns is two.

BloodstainPatternAnalysis-ANewChallengeforComputationalIntelligenceCommunity

215

2 3 4 5 6 7 8 9

0.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

number of clusters

overall average silhouette width

Figure 9: The overall average silhouette widths for each

tested number of clusters k.

Once identiﬁed the number of blood spatter patterns

in a photograph, it is necessary to map the blood-

stains to one of the identiﬁed patterns. This mapping

is computed by considering the membership degrees

of the pairwise points of convergence with respect to

the computed clusters. In detail, the proposed ap-

proach assumes that a bloodstain belongs to the pat-

tern/cluster for which the membership degrees of the

pairwise points of convergence involving the blood-

stain are higher. Precisely, for each bloodstain, a so-

called pattern membership value is computed for each

identiﬁed pattern/cluster by performing the mean of

the membership degrees of pairwise points of conver-

gence involving the bloodstain and belonging to the

considered cluster. At the end, the bloodstain belongs

to the cluster/pattern for which the computed pattern

membership value is the highest. Once identiﬁed the

blood spatter patterns and the relative bloodstains, the

pairwise points of convergence involving bloodstains

belonging to different patterns are removed from the

set F since they are considered outliers. At this mo-

ment, the proposed system computes the center of

the updated clusters (changed by removing outliers)

which represents the point of convergence of the rel-

ative blood spatter pattern. By using the point of con-

vergence and the angles of impact of each bloodstain

belonging to the pattern, it possible to compute the

point of origin of each pattern as described in section

4 CONCLUSIONS

This position paper introduces a novel application do-

main in the area of computational intelligence: BPA.

As proved by a preliminary study, fuzzy reasoning

can strongly improve the capabilities of investigators

in addressing BPA issues and try to solve complex

14 16 18 20 22 24

8.5

9.5

10.5

11.5

12.5

Figure 10: Output of Fuzzy C-Means for the number of

clusters equals to 3.

cases in a faster and more precise way than current

manual techniques. Our opinion is that BPA could

represent a breakthroughapplication in computational

intelligence community and open new research sce-

narios in a novel challenging area such as the Forensic

Intelligence.

REFERENCES

Bezdek, J. C. (1981). Pattern Recognition with Fuzzy Ob-

jective Function Algorithms. Kluwer Academic Pub-

lishers, Norwell, MA, USA.

Boonkhong, K., Karnjanadecha, M., and Aiyarak, P. (2010).

Impact angle analysis of bloodstains using a simple

image processing technique. Sonklanakarin Journal

of Science and Technology, 32(2):169.

Brodbeck, S. (2012). Introduction to bloodstain pattern

analysis. Practice (Vol. 2), 51:57.

Fitzgibbon, A., Pilu, M., and Fisher, R. (1999). Direct least

square ﬁtting of ellipses. Pattern Analysis and Ma-

chine Intelligence, IEEE Transactions on, 21(5):476–

480.

James, S. H., Kish, P. E., and Sutton, T. P. (2005). Principles

of bloodstain pattern analysis: theory and practice.

CRC Press.

Rousseeuw, P. J. (1987). Silhouettes: A graphical aid to

the interpretation and validation of cluster analysis.

Journal of Computational and Applied Mathematics,

20(0):53 – 65.

Shen, A. R., Brostow, G. J., and Cipolla, R. (2006). Toward

automatic blood spatter analysis in crime scenes. In

Crime and Security, 2006. The Institution of Engineer-

ing and Technology Conference on, pages 378–383.

IET.

FCTA2014-InternationalConferenceonFuzzyComputationTheoryandApplications

216