Process Management Enhancement by using Image Mining Techniques:

A Position Paper

Myriel Fichtner

, Stefan Sch

onig

and Stefan Jablonski

University of Bayreuth, Germany

University of Regensburg, Germany

Keywords:

Image Mining, Process Model Enhancement, Quality Control, Recommendation System, Process Redesign.

Abstract:

Business process modeling is a well-established method to deﬁne and visualize business processes. In complex

processes, related process models may become large and hard to trace. To keep the readability of process

models, process details are omitted. In other cases, process designers are not aware which process steps should

be modelled in detail. However, the input speciﬁcation of some process steps or the order of internal sub-steps

could have an impact on the success of the overall process. The most straightforward solution is to identify the

cause of reduced process success in order to improve the process results. This can be challenging, especially in

ﬂexible process environments with multiple process participants. In this paper we tackle this problem through

recording image data of process executions and analyzing them with image mining techniques. We propose to

redesign business process models considering the analysis results to reach more effective and efﬁcient process

executions.

1 INTRODUCTION

Companies use business process models to visualize

and control internal workﬂows describing the nec-

essary process steps for each process participant to

reach a company goal, for example manufacturing

a certain product. Such process models may con-

tain hundreds of modeling elements leading to large

and hardly traceable process models (A. Polyvyanyy,

2008b). Besides, though there exists modeling rec-

ommendations regarding general aspects like correct-

ness or comparability (J. Becker, 2000), there is no

concrete rule or guidance how detailed a process has

to be modelled. Thus, it is quite obvious that pro-

cesses are often modelled in an abstract way to keep

up clarity and traceability. Sometimes also missing

knowledge about process details prevents a process

modeller to add more detailed (sub-)steps. This high

level of abstraction leads to the fact that detailed infor-

mation of process steps are omitted (R. Bobrik, 2007),

(A. Polyvyanyy, 2008a). Two typical examples illus-

trate the observation from above.

In a ﬁrst case, diverse liquids have to be ﬁlled into

a casting mold. Due to missing knowledge or due

to keeping the process description simple, the pro-

cess modelers describe this process step on an abstract

level as “add all ingredients”. Looking into process

details reveals that three ingredients have to be added.

So the process could also be reﬁned into three process

(sub-)steps ”add blue liquid”, ”add red liquid”, and

”add green liquid”. Process participants are free to

choose the execution order of low-level process steps

which are not modelled but contained implicitly, lead-

ing to an excessive ﬂexibility in the execution step.

In some other cases, process modelers are not aware

of the optimal execution sequence of process steps.

For instance, they allow that two steps are executed

independent from each other, i.e. in arbitrary order.

However, the execution sequence does have an im-

pact on process performance, regarding the execution

time and/or the quality of process outcomes. Follow-

ing the example above, the addition of the red liquid

as second step instead of third step results in a mixture

with reduced binding capability, i.e. this execution se-

quence is ﬁnally not desired. This affects the overall

process success since the result of any subsequent task

depends on the quality of the mixture.

In a second case, parts have to be disposed on a

pallet. Again, due to missing knowledge or due to

keeping the process description simple, the process

designers are not more speciﬁc about placing the parts

on the pallet. Furthermore, some tasks are hard to de-

scribe or important details can’t be described with es-

tablished process modeling languages. In this palleti-

Fichtner, M., Schönig, S. and Jablonski, S.

Process Management Enhancement by using Image Mining Techniques: A Position Paper.

DOI: 10.5220/0009573502490255

In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 1, pages 249-255

ISBN: 978-989-758-423-7

249

sation scenario this may include (i) information con-

sidering the position of objects in the product environ-

ment, (ii) information that is related to special move-

ments. If modellable, such information can only be

integrated in a process model through a large num-

ber of process modeling elements. Especially if dif-

ferent alternatives are allowed, the number of mod-

eling elements strongly increases what contradicts

the ideal to keep business process models traceable

through preserving a certain level of abstraction. Al-

though there are approaches that deal with that chal-

lenge, e.g. (Wiedmann, 2017), (M. La Rosa, 2011),

(A. Polyvyanyy, 2008b), they assume that knowledge

about the correlation between this details and the pro-

cess success already exists. To guarantee the success

of a process and to enhance its execution, necessary

details of activities that are not contained in the busi-

ness process model have to be discovered while satis-

fying the following three conditions:

1) Process analysis has to be done automatically since

participants are not aware that execution details may

inﬂuence the process success. Also in the case that the

process model is executed for the ﬁrst time, no prior

or expert knowledge exists. Furthermore, the analysis

should be done with a minimum interference in the

regular working process to ensure correct results.

2) The analysis should be able to identify the cause of

reduced process success by extracting necessary in-

formation and developping suggestions for improve-

ment.

3) The extracted information has to be represented in

the process model in an appropriate way.

Although quality control and process monitoring

are popular topics in research, existing approaches

(e.g. (D. E. Lee, 2006), (T. Pryk

ari, 2010)) focus on

the identiﬁcation of deviations in process results but

do not meet the listed requirements. In this paper we

suggest to tackle this problem by recording and ana-

lyzing image data of activity executions. We propose

to apply image mining techniques in the process man-

agement context to enhance business process models

and ensure process sucess.

In our work, we extract necessary process steps

and related information from image data which is not

yet contained in the existing process model. In or-

der to identify these missing details, we propose to

use established image mining techniques. The pro-

cess model is then enriched by this content through

adequate techniques. We therefore go beyond the lim-

itations of regular process modeling languages.

In general camera systems are cost-efﬁcient

sensors which often already exist in small- and

medium-sized enterprises and industrial working ar-

eas. Recorded image data created in such process en-

vironments contains valuable information which is of-

ten not fully analyzed. Image data can either contain

static information, like the input or output of a task or

dynamic information, like the task processing itself.

Both may contain necessary information for process

enhancement. In our conceptual approach, we there-

fore dinstinguish between the analysis of images and

videos. Images are related to snap-shots of the execu-

tion while videos capture the whole execution of the

task including related subtasks.

Our overall system meets all three requirements as

described above. Furthermore, the image analysis re-

sults of our conceptual approach may serve as input

for recommendation systems to support given recom-

mendations. For example, our system reveals the best

execution of an activity and therefore identiﬁes the

most suitable process participant for this task. In con-

trast to previous work that relates images to process

context, we focus on the process execution step and

analyze recorded image data that contains real infor-

mation of the process environment.

The remainder of this paper is structured as fol-

lows. The following section summarizes background

information and gives an overview of important re-

lated work. Section 3 presents our conceptual ap-

proach to reach process enhancement through analyz-

ing image data and points out our contribution. We

conclude our work and give a recommendation for fu-

ture work in Section 4.

2 BACKGROUND AND RELATED

WORK

In general, images can be understood as complex

data collection. Depending on the context in which

they are created, the knowledge about this context

and other associations, images may contain meaning-

ful information if analyzed and interpreted correctly.

How to achieve an effective extraction of this infor-

mation is the research question in work that is re-

lated to image mining. According to (J. Zhang, 2001),

image mining deals with the extraction of implicit

knowledge, image data relationship, or other patterns

not explicitly stored in the image databases. Among

others, different methods from computer vision, data

mining and machine learning are used to process low-

level pixel representations contained in raw images or

image sequences in order to identify high-level spatial

objects and relationships. The overall image mining

process is well described in (M. Hsu, 2010). Summa-

rized, the process can be divided into three parts:

Pre-processing: In order to reduce the cost of the

analysis step which can be high in time and space,

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

250

images have to be preprocessed. Therefore unneces-

sary or unrelated data is cleaned up and quality reduc-

tions due to noise are eliminated through ﬁltering op-

erations. This step may include image-thresholding,

border-tracing an wavelet-based segmentation.

Feature Extraction: Algorithms are used to detect

features such as shapes, edges or other basic elements

in the images. Therefore the image content is reduced

while unimportant features can easily be discarded. A

promising feature extraction approach can be found in

(P. G. Foschi, 2002), where a combination of the fea-

tures color, edge and texture is suggested.

Image Mining Technology: Image mining tech-

niques are used on the extracted feature vectors to

reveal, evaluate and explain high-level knowledge.

Several methods have been developped which real-

ize this procedure in different ways: Image classiﬁca-

tion, clustering, indexing and retrieval, object recog-

nition, association rule mining and approaches that

work with neural networks.

The techniques are used in many different real-

world applications, like for example the analysis of

paths and trends of forest ﬁres over years in satel-

lite images in order to enable ﬁreﬁghters to ﬁght ﬁre

more effectively (J. Zhang, 2002). Another work

uses images gathered from the Web for learning of

a generic image classiﬁcation system and enables a

Web image mining approach for generic image clas-

siﬁcation (Yanai, 2003). Image mining was intro-

duced by (Ordonez and Omiecinski, 1998) as new ap-

proach for data mining. The fundamental concepts of

discovering knowledge from data stored in relational

databases are transfered to image databases.

Related to this idea and the fact that process min-

ing builds on data mining, also the association of im-

age mining with process mining techniques or rather

the application of image mining in the context of busi-

ness process management is reasonable. However,

the application of image mining techniques or in gen-

eral the integration of images in this ﬁeld is not yet

fully explored. The work of (Wiedmann, 2017) and

(R. Schmidt, 2016) suggest two different approaches

to introduce images and related mining techniques in

business process management. In the thesis of (Wied-

mann, 2017), the business process modeling language

BPMN is extended to a more intuitive modeling lan-

guage which allows to annotate tasks with multime-

dial content like images or videos. This approach en-

ables to add non-formalized descriptions to a process

task and enriches the process model with additional

information. During the execution of this task, pro-

cess participants can follow the referenced execution

in the video. The work of (R. Schmidt, 2016) con-

ﬁrms the potential of image mining for business pro-

cess management. The process management lifecycle

according to (M. Dumas, 2013) is presented and the

application and intregation of images and suitable im-

age mining techniques for each phase are discussed.

This approach focuses on image data which is cre-

ated in each phase. The authors differentiate between

documents, drawings and pictures, while documents

contain textual information and are analyzed with op-

tical character recognition methods. Drawings and

pictures are analyzed by using one of the image min-

ing techniques as described above. Furthermore, the

authors present a protoype for object recognition of

business process models which detects modeling ele-

ments like gateways, activities etc. from images.

Altogether, both authors explain how images con-

tribute to support the overall process. Particulary in-

teresting is the suggestion of (R. Schmidt, 2016) to

analyze pictures, taken with phones or tablets, which

contain information of the production environment.

Covering the process monitoring and controlling step,

these images are analyzed subsequently in order to

ﬁnd possible process improvements. This proposal

corresponds with our approach, while we share the

idea that any issues that may reduce the overall pro-

cess success could revealed through monitoring the

process execution.

In contrast to the work of (R. Schmidt, 2016), we

suggest a concrete approach that aims to process im-

provement through an overall system that solely bases

on the image data that is produced in the process ex-

ecution step (cf. Section 3). Our system is restricted

to images or videos which contain real information of

the production environment or capture actual states of

a product. We go further by restoring the analyzed im-

provements in the existing process model. For this we

propose to translate them in the considered process

modeling language or to use media annotations like

suggested by (Wiedmann, 2017). Compared to clas-

sical machine vision approaches that control the qual-

ity of a product like (Manigel and Leonhard, 1992),

(H. Paulo, 2002) or (E. Saldana, 2013), our concept

starts one stage earlier and identiﬁes the causes of de-

fects in products if they are related to human task ex-

ecutions. However, such systems can easily be inte-

grated in our overall concept while the implemented

techniques can be used in the image analysis step of

our concept.

3 CONCEPTUAL APPROACH

We illustrate our conceptual approach by a running

example. In this example, a certain product has to be

manufactured according to a process model PM. We

Process Management Enhancement by using Image Mining Techniques: A Position Paper

251

Figure 1: Our overall concepts consists of 5 high-level steps while in one cycle, task t

of all process tasks T is taken into

account. Furthermore, the existing business process model PM is adapted to PM

and the database DB serves as storage

for pairs of recorded image data (images and videos) and related feedback of each process execution (i, f ). Based on the

validation of the calculated reference data r, either PM is extended with missing information to PM∗, or new reference data

is evaluated or t

n+1

∈ T is taken into account.

assume that PM contains all tasks T of the process

necessary to create that product. Although all (most)

process executions were successful, i.e. the manufac-

turing processes were completed correctly, the ﬁnal

products were revealing quite different quality. Since

the process model is already in place and process ex-

ecutions did not show errors, the assumption was that

not all process steps were modelled – and thus ﬁnally

performed – optimally. The process experts presumed

that either (i) some of the process steps were not mod-

elled in enough detail, i.e. are too abstract since in-

ternal sub-steps are not speciﬁed sufﬁciently (process

steps of Type A); or that (ii) the input speciﬁcation

of some process steps were not modelled prescrip-

tively enough (process steps of Type B). Focusing

these two ﬂaws in this paper, our conceptual approach

comprises the following 5 steps (cf. Figure 1).

1. Preparation.

By examining the process model, the process

experts identify all tasks of Type A and Type B

as candidate steps that might lead to disparate

production results. Although, at this point in time,

the process experts still do not know whether this

assumption is correct and in case it is correct

how the solutions would look like. All tasks that

meet these conditions are summarized in a task

list T

⊆ T . To continue our overall example, we

assume that we analyzed PM while identifying

exactly two tasks t

(”add all ingredients to

the glass”) matching Type A and t

(”place all

machine parts on the palette”) matching Type B

with t

, t

∈ T out of overall n tasks (|T | = n)

leading to T

= {t

, t

2. Deﬁnition.

In this step, all tasks contained in T

are analyzed

incrementally. That means that any task t ∈ T

is selected and prepared for further examination.

Therefore PM has to be redesigned to PM

while

the deﬁnition of t has to be adapted as follows. If

t belongs to Type A, the new deﬁnition of this task

has to contain the information that its execution

has to be monitored by recording videos of the ex-

ecution. If t belongs to Type B, the task has to be

redesigned so that an image has to be taken. De-

pending on the context, the image has to be taken

at the beginning, before starting the execution or

after ﬁnishing the execution of t. Furthermore, a

camera system has to be provided and installed in

order to enable the recording of image data. In our

example, the process experts perceive that task t

is composite. However, they are not sure what

the sub-tasks are at all. So, they should moni-

tor the upcoming executions of this process step

by videoing it. Additionally, the process experts

see that task t

requires a complex input conﬁg-

uration. Thus, it is proposed that process experts

take pictures of the input conﬁguration of upcom-

ing process executions. Due to the presumption of

the process experts that t

might have more impact

on the overall process sucess, they decide to start

with the analysis of t

. Therefore PM is redeﬁned

in the way that an image is taken after placing all

machine parts on the palette.

3. Execution and Labelling.

In this step, the redeﬁned process model PM

executed and image data (images and videos) is

generated as speciﬁed in the model. At the end

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

252

of each process execution, the data is labelled,

i.e. process experts evaluate the process success

through giving feedback considering different cri-

teria. Traditional criteria that are related to pro-

cess success are the production time, cost and

quality of the result (A. Collins, 2004). We sug-

gest to implement the feedback step at the end of

the overall process execution in order to identify

the inﬂuence of the considered task on subsequent

tasks in the process model. Depending on the

situation and the issue, giving feedback directly

after the execution of the considered task possi-

bly does not reveal correlations within the process

like we will see later in the example. However, if

the feedback is given directly after the execution,

the considered task is analyzed related to classi-

cal machine vision approaches but provides more

ﬂexibility. For both alternatives it should be en-

sured that all feedback refers to the same criterion

in order to enable comparability between them.

Any feedback f and the related image data i that

has been recorded during the execution step are

stored as pair (i, f ) in a database DB. Each set

of image data referring to one execution is hence

associated with an evaluation of success. This

procedure is repeated until DB contains enough

data to serve as basis for a meaningful analysis.

In our running example, we assume that PM

executed 100 times under the assumption that we

know that 100 executions provide a sufﬁcient data

basis for analysis. This means that 100 images are

recorded, which show the resulting palette after an

employee placed all machine parts on it. Together

with the associated feedback 100 pairs (i, f ) are

stored in DB.

4. Image (and Video) Analysis.

In this step, all entries in DB are analyzed by us-

ing image mining techniques including all 3 parts

of the image mining process (cf. Section 2). De-

pending on the type of i that is stored in DB, dif-

ferent objectives are deﬁned and therefore differ-

ent methods have to be applied. In all cases, the

overall goal is to analyze the differences between

all entries and to relate them with the feedback in

order to identify which differences cause good or

insufﬁcient process results. Based on the analysis

results, reference data r is created which serves as

input for the validation step and the camera sys-

tem. This data can be understood as the image

data which is recorded in the context of an ideal

execution and is used as guideline for further exe-

cutions.

If DB contains images (i.e. i is from Type A), it

holds entries capturing the same state of each pro-

cess execution. To somehow compare these im-

ages and to ﬁnd relevant features that differ them,

image retrieval techniques as well as image clas-

siﬁcation and image clustering are suitable meth-

ods. But also the application of neural networks is

quite promising, especially in ﬁnding an adequate

reference image. The most promising image, i.e.

the arrangement that is rated best after feedback

evaluation, is stored as guideline in r.

If DB contains videos (i.e. i is from Type B), it

holds several recordings of the full execution of

the same task. As described above these execu-

tions may differ in the order of underlying sub-

tasks. Therefore all subtasks has to be identiﬁed

in a ﬁrst step and then the order of them has to be

analyzed and compared in a second step. These

requirements could be reformulated as challenge

to extract event logs and therefore process mod-

els from videos and to ﬁnd techniques to compare

them. The order of the subtasks that has the most

success is ﬁnally stored in r.

In our example, the 100 entries in DB are ana-

lyzed, while it turns out that the positioning of the

machine parts on the palette seems to have an im-

pact on the overall process sucess. The image data

which refers to an ideal execution is stored in r

and is prepared as input for the camera system.

5. Validation.

The instructions identiﬁed in phase ”Image Anal-

ysis” are incorporated into PM

resulting in PM∗,

while t is modiﬁed to match r. Therefore PM can

be extended by additional modeling elements to

include necessary subtasks or other detailed in-

formation. Alternatively we suggest to follow the

concept of (Wiedmann, 2017) and to use media

annotations. This approach enables the integra-

tion of image data and allows the use of a more

powerful process modeling language what supp-

ports our idea of process enhancement.

Afterwards, this new process model version PM∗

is further executed. Observing and measuring

process success determines whether this new ver-

sion of a process is accepted or further investiga-

tion have to be undertaken. In the latter case, the

whole process improvement process has to start

from its beginning.

To realize this procedure, t is still taken into ac-

count and image data is recorded while PM∗ is

executed. In contrast to the previous execution

step related to PM or PM

, process participants

now execute an estimated more successful version

of t. Furthermore, the camera systems recognize

deviations, leading to an interruption of the exe-

cution if the recorded image data differs from r.

Process Management Enhancement by using Image Mining Techniques: A Position Paper

253

Figure 2: The image i recorded after executing a task (left), the reference image r (middle) and the calculated difference image

(right). Black areas in r refer to pixel colors that deviate from each other, while white areas indicate same pixel colors.

If it does, the execution has to be adapted until

it matches r including a predeﬁned threshold. In

this step, again image mining techniques are used

to identify this deviations. Just as in the previ-

ous execution step, pairs of image data and feed-

back are collected and analyzed in order to vali-

date if the restriction to executions that relate to

r really lead to process sucess. If the validation

fails and DB still contains negative feedback, ei-

ther r has to be recomputed based on a larger num-

ber of entries in DB or t was not the task that led

to the unsatisfying process success and the pro-

cedure has to be repeated for the next task t

∈ T ∗

while T∗ = T

\{t}. If the deﬁnition of r as guide-

line leads to more successful process executions,

the modiﬁed process model PM∗ replaces PM.

Applied to our example, we assume that the pro-

cess model is adapted as described above. We fur-

ther assume that during the execution of t

, de-

viations are detected because a machine part was

placed wrong on the palette. For this we imple-

mented a prototype and built up a small scenario

according to t

as shown in Figure 2. While there

exist several approaches to ﬁnd the difference in

two frames, one of the easiest ways is to determine

the pixel differences. In this technique the num-

ber of pixels that change in value more than some

threshold are counted (cf. (Boreczky, 2006)). By

comparing i and r it can be seen that two of three

machine parts differ in their position and orienta-

tion leading to their reoccurences in the difference

image. In this example we assume that the valida-

tion step fails. The process experts decide to fur-

ther examine t

instead of taking t

into account.

Therefore further process model executions are

performed in order to reach a larger data set for

DB and to identify a better reference image r

The presented example shows how our concept can

be applied to enhance processes through using image

mining techniques. Our approach enables the identi-

ﬁcation of process model execution details that have

an impact on the overall success of the process. The

existing process model is only extended by informa-

tion, which is identiﬁed as necessary in the analysis

step. Therefore the predeﬁned abstraction of the pro-

cess model is preserved but details which improve the

overall process are included.

Since the presented idea considers the overall success,

it is possible to reveal complex interrelations and de-

pendencies between tasks that are not contained in the

process model but inﬂuence the process result. Qual-

ity assurance approaches only evaluate the result of

a single (mostly non-human) task execution. There-

fore they are only able to analyze the direct correla-

tion of a task execution with its output. In contrast, we

suggest to consider the overall process success and to

focus on all tasks, where the input is not ﬁxed or mul-

tiple executions are possible. Our concept therefore

enables to analyze the effect of a task on a subsequent

task which leads to a reduction of the process suc-

cess. In most cases, all single tasks were executed

according to their rules deﬁned in the process model

while an direct evaluation of the task execution result

would conﬁrm process success. In contrast, the re-

sult of the overall process might not be satisfying. At

this point, existing approaches reach their limitations

since they are not able to identify the cause of the re-

duced process success sourcing in the dependency be-

tween tasks. Like described above, our concept tack-

les this issue and provides a more general process ex-

ecution analysis.

Our approach is based on the analysis of image data

while we therefore introduce an innovative appli-

cation of image mining techniques in the process

management context. Image data serves as valuable

source for complex analysis since it has high informa-

tion content, provides several features and supports

the ﬂexibility of our approach. This ﬂexibility reﬂects

in the possibility to deal with static as well as dynamic

information related to the process environment.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

254

Concluding, our system does not only extract relevant

features from images but identiﬁes process execution

dependencies affecting the process success and sug-

gests reference data related to optimal executions. We

want to point out that the intregration of this infor-

mation in the existing process model is important to

ensure process success in future executions, while we

suggest media annotations presented by (Wiedmann,

2017) as adequate technique.

4 CONCLUSION AND FUTURE

WORK

In this paper we present our idea to use image min-

ing techniques in order to enhance business pro-

cesses. We suggest a ﬂexible approach which is

based on monitoring process executions while image

data related to the process environment is collected.

This image data is labelled and analyzed to identify

execution-speciﬁc features that have an impact on the

overall process sucess. We further show how this in-

formation can be integrated in existing process mod-

els to enhance future executions without reducing the

traceability of process models. The review of exist-

ing approaches conﬁrms that the application of image

mining techniques in the process management context

is an open research gap.

As for future work, technical aspects have to be

discussed which include the evaluation of existing im-

age mining techniques depending on the use case.

Furthermore, we plan to implement an overall pro-

totype and to proceed evaluations with simple exam-

ples. Finally, we intend to further investigate the real-

ization of our idea in real process environments. We

will collect broader use case scenarios and explore

suitable application areas.

REFERENCES

A. Collins, D. B. (2004). Project success - a survey. In

Journal of Construction Research, volume 5, pages

211–231.

A. Polyvyanyy, S. Smirnov, M. W. (2008a). Process model

abstraction: A slider approach. In 2008 12th Interna-

tional IEEE Enterprise Distributed Object Computing

Conference.

A. Polyvyanyy, S. Smirnov, M. W. (2008b). Reducing com-

plexity of large epcs. In EPK 2008 GI-Workshop,

Saarbr

ucken.

Boreczky, J. (2006). Comparison of video shot boundary

detection techniques. In Journal of Electronic Imag-

ing, volume 5, pages 122–128.

D. E. Lee, I. Hwang, C. V. (2006). Precision manufacturing

process monitoring with acoustic emission. In Condi-

tion Monitoring and Control for Intelligent Manufac-

turing.

E. Saldana, R. Siche, M. L. (2013). Review: computer vi-

sion applied to the inspection and quality control of

fruits and vegetables. In Brazilian Journal of Food

Technology.

H. Paulo, R. Davies, B. C. (2002). A machine vision quality

control system for industrial acrylic ﬁbre production.

In EURASIP Journal on Advances in Signal Process-

ing 2002.7.

J. Becker, M. Rosemann, C. V. U. (2000). Guidelines of

business process modeling. In Business process man-

agement.

J. Zhang, W. Hsu, M. L. (2001). Image mining: Issues,

frameworks and techniques. In Proceedings of the

2nd ACM SIGKDD International Workshop on Mul-

timedia Data Mining (MDM/KDD’01), pages 1–7.

J. Zhang, W. Hsu, M. L. (2002). Image mining: Trends and

developments. In Journal of Intelligent Information

Systems, volume 19, pages 7–23.

M. Dumas, M. La Rosa, J. M. (2013). Fundamentals of

Business Process Management. Springer, Heidelberg.

M. Hsu, S. Y. (2010). Overview of image mining re-

search. In ICCSE 2010 - 5th International Conference

on Computer Science and Education, Final Program

and Book of Abstracts, pages 1868–1970.

M. La Rosa, A. H. Ter Hofstede, P. W. (2011). Managing

process model complexity via concrete syntax modiﬁ-

cations. In IEEE Transactions on Industrial Informat-

ics.

Manigel, J. and Leonhard, W. (1992). Vehicle control by

computer vision. In IEEE Transactions on industrial

electronics 39.3.

Ordonez, C. and Omiecinski, E. R. (1998). Image mining:

A new approach for data mining. Technical report,

Georgia Institute of Technology.

P. G. Foschi, D. Kolippakkam, H. L. (2002). Feature ex-

traction for image mining. In Multimedia Information

Systems, pages 103–109.

R. Bobrik, M. Reichert, T. B. (2007). View-based process

visualization. In International Conference on Busi-

ness Process Management.

R. Schmidt, M. Mohring, A. Z. (2016). Potentials of image

mining for business process management. In Intelli-

gent Decision Technologies 2016. Springer.

T. Pryk

ari, J. Czajkowski, E. A. (2010). Optical coher-

ence tomography as an accurate inspection and qual-

ity evaluation technique in paper industry. In Optical

review.

Wiedmann, P. (2017). Agiles Gesch

aftsprozessmanagement

auf Basis gebrauchssprachlicher Modellierung (Doc-

toral dissertation).

Yanai, K. (2003). Web image mining toward generic image

recognition. In Poster Proceedings. 12th International

World Wide Web Conference, pages 1–6.

Process Management Enhancement by using Image Mining Techniques: A Position Paper

255