A Service-Based Preset Recommendation System for Image Stylization
Applications
F. Fregien
1
, F. Galandi
1
, M. Reimann
1 a
, S. Pasewaldt
2
, J. D
¨
ollner
1
and M. Trapp
1 b
1
Hasso Plattner Institute, University of Potsdam, Prof.-Dr.-Helmert-Str. 2-3, 14482 Potsdam, Germany
2
Digital Masterpieces GmbH, August-Bebel-Str. 26-53, 14482 Potsdam, Germany
fl
Keywords:
Image Stylization, Recommendation System, Microservice.
Abstract:
More and more people are using images and videos as a communication tool. Often, such visual media is
edited or stylized using software applications to become more visually attractive. The data that is produced
by the editing process contains useful information on how users interact with the software and data yielding
respective results. In this context, this paper presents a framework that facilitates data storage, data profiling,
and data analysis of image-stylization operations, image descriptors, and equivalent usage data by means of a
recommendation system. The presented concept is implemented prototypical and preliminary evaluated.
1 INTRODUCTION
Motivation. In recent years, the number of smart
phone users increased continuously. With the num-
ber of mobile users is growing, so is the amount of
visual media data, such as images or videos. Since
usually every smartphone has a camera, there exist
several million professional and hobby photographers
around the world. To transform, enhance, or stylize
photos, there are a variety of editing applications that
enable applying filters or adjust, e.g., the contrast or
brightness of an image. While highly-skilled users re-
quire low-level control over all filter options, a com-
mon user can be overwhelmed by the big amount of
configuration options (Figure 1). It requires a time-
consuming learning phase or trial and error to achieve
satisfying results. For casual users, this can be a frus-
trating experience as they expect pleasant results fast.
To overcome this issue, we utilize an approach
known as recommendation system. Recommendation
Systems (RSs) is an active research field in the area
of Machine Learning (ML). It has been widely ap-
plied in e-commerce and other online services used
on an every-day basis, e.g., when listening to mu-
sic (Moscato et al., 2021), watching movies (Kumar
et al., 2015), buying products (Aciar et al., 2006), or
a
https://orcid.org/0000-0003-2146-4229
b
https://orcid.org/0000-0003-3861-5759
being on social media (Sperl
´
ı et al., 2018). The goal
of an RS is to predict users’ preference and recom-
mend the most relevant items to users. In order to
provide content-based recommendations, the system
must contain information about the users and items.
Further, understanding the raw data collected by
a recommendation system and forming knowledge,
is advantageous for a company or developers seek-
ing to understand its users. For example, knowing
what users like and how they interact with a piece of
software is vital to adjust features and release plan-
ning. Furthermore, usage data enables the possibility
to learn from expert users to eventually transfer useful
insights like parameter settings for less professional
users. A recommendation system is a useful tool to
hide complex computations and settings to increase
user experience and satisfaction. Such recommenda-
tion systems enable the implementation of so-called
“one-click solutions” for choosing image filters based
on user preferences or the image content. To enable
such systems, a data storage framework has to facili-
tate data profiling and analysis.
Problem Statement & Challenges. To summarize
the above, state-of-the-art in (mobile) image-filtering
apps provide a high number of available filter op-
erations that can be combined in various configu-
rations, yielding numerous output possibilities (Fig-
Fregien, F., Galandi, F., Reimann, M., Pasewaldt, S., Döllner, J. and Trapp, M.
A Service-Based Preset Recommendation System for Image Stylization Applications.
DOI: 10.5220/0011772800003417
In Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 2: HUCAPP, pages
251-258
ISBN: 978-989-758-634-7; ISSN: 2184-4321
Copyright
c
 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
251

Figure 1: Different configurations of image-stylization
techniques allow graphically different results. These can
generate numerous variants for an input image. The vari-
ants shown here were obtained by manually adjusting the
process parameters.
ure 1). According to Isenberg, each individual opera-
tion is usually highly-configurable at different level-
of-control (Isenberg, 2016): from selecting a num-
ber of customizable global presets, over controlling
a number of global parameter values, to local adjust-
ments (D
¨
urschmid et al., 2017). While preset facili-
tate ease-of-use, the latter two enable a low level-of-
control featuring highly individual results and styles.
However, they also lead to increased operating com-
plexity and often require time-consuming trial and er-
ror methods. With respect to this, major challenges
represent the design and implementation of a system
that enables the data collection and analysis as well as
the provisioning of recommendations based on user
characteristics and specifics inputs.
Approach & Contribution. To address the chal-
lenges above, this paper describes the architectural
design and implementation of a microservice-based
framework to store and retrieve usage data as well
as a prototype of a recommendation system that pro-
poses most used image operations or operation pre-
sets. To summarize, the paper makes the following
contributions. It presents the concept of a recommen-
dation system for image-stylization applications and
it describes a prototypical implementation based on
a microservice architecture and framework. The re-
mainder of this paper is structured as follows: Sec-
tion 2 reviews some terminology and background in
the context of microservice-based image processing
as well as basic approaches and challenges for de-
signing recommendation systems. In Section 3, we
describe the workflow and implementation of how to
collect, store, and analyze usage data. In Section 5,
we evaluate the runtime performance of our system
by means of an application example, and present as-
pects of future work. Section 6 concludes this paper.
2 BACKGROUND
This section briefly describes related and previous
work on the topic of microservices for image process-
ing, approaches for the collection and provisioning of
user or usage data, and fundamentals of recommen-
dation systems.
Microservices for Image Processing. In recent
years, microservices have become increasingly pop-
ular. Unlike a monolithic system, microservices are
autonomous modules that perform various tasks with
respect to the application logic. The advantages of
microservices are (1) scalability of the components,
(2) easier deployment and maintainability as well as,
(3) the possibility to introduce various technologies
into one system (Viggiato et al., 2018). Despite the
advantages, a microservice infrastructure introduces
an information barrier between the services, that can
have negative effects on system latency and perfor-
mance (Shadija et al., 2017).
In our work, we are extending an existing mi-
croservice platform for cloud-based visual analysis
and processing, which was first presented by Richter
et al. (Richter et al., 2018). The microservice plat-
form has been improved and further developed in sev-
eral subsequent works. The core of this platform is
the Image Processor service that applies Operations
to images. The term “operations” refers to image fil-
ters that change the visual appearance as well as trans-
formations, such as rotating, mirroring, or cropping.
To even realize complex filters that, e.g., make use
of Neural Style Transfer (NST) (Gatys et al., 2016),
the Image Processor uses so-called Visual Comput-
ing Assets (VCAs) (D
¨
urschmid et al., 2017). They
are a textual representation, which allow to describe
programmatically how an operation is applied on the
Graphics Processing Unit (GPU) using shader pro-
grams. The specific results of an operation can be
adjusted by using Parameters. A set of parameters
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
252

and an assignment of parameter values compose to
a Preset, which are used to label specific parameter
configurations that are proven to be aesthetic. Op-
erations can even be chained in a Pipeline to create
more advanced filters. Using Visual Computing As-
sets (VCAs), we are able to profile and analyze which
operations pipelines, presets, or parameters are used
most frequently.
Collecting, Profiling, and Analyzing Data. Data
profiling is an important part of any data-collecting
mechanism. By understanding the dataset and its
metadata, the underlying database schema can be im-
proved, design flaws can be detected and the data
can be better prepared for further analyzing such as
machine learning tasks (Abedjan et al., 2017). Data
analysis helps to extract useful information from data
and has a wide range of applications. Data mining
as a method of data analysis is revealing interesting
patterns in the data that can help improving business
workflows, finding new scientific discoveries, or en-
hancing user experience (Han et al., 2011). Especially
in the business domain, data analysis supporting bet-
ter decision-making and showing improvements in
the business logic (Xia and Gong, 2014). Related to
web applications, collecting usage data can help im-
proving user experience and usability of the applica-
tion (Beri and Singh, 2013).
In our use case, we focus on analyzing opera-
tions, their presets and parameters, as well as how
they are arranged in a pipeline by a user. Further-
more, we collect image descriptors, which contain
several metadata about an image, such as Exchange-
able Image File Format (EXIF) or Global Position-
ing System (GPS) data (
¨
Olveck
´
y and Host’oveck
´
y,
2021). With these, the recommendation system can
try to find correlations between operation configu-
rations and images. To persist the collected usage
data, we use a state-of-the-art storage approach: a
Graph-based Database Management System (GDMS)
(Huang et al., 2002). We decide to use Neo4j, as it
is open-source and self-hostable, yet it is one of the
most stable and mature GDMS that provides methods
to run complex queries on related data.
Recommendation Systems. A recommender or
recommendation system is suggesting solutions or
proposing hints to solve a particular problem of a
user (Ricci et al., 2022). Users often do not have the
capability to make respective decisions because they
are hindered by time constraints or lack of operating
knowledge. Therefore, recommendation systems are
useful tools to manage a large number of options and
are supporting the decision-making process. Due to
the advantages for the user, recommendation systems
are heavily used in commercial applications to in-
crease user satisfaction and sales (Ricci et al., 2010).
There are three major types of RSs: (1) content-
based RSs analyze the content of items to recommend
items that are similar to the ones the user liked in
the past, (2) Collaborative Filtering RSs makes pre-
dictions based on the preferences of other users with
similar tastes, and (3) Hybrid Recommendation RSs
that combine collaborative filtering and content-based
methods to leverage the advantages of both.
Besides the commonplace “customers also
bought, used or watched” recommendation, more
complex systems are developed. Novel step-wise
recommendations are designed to support users
accomplishing tasks that consist of several steps
instead of one (Nouri et al., 2020). This scenario
resembles the decision-making process of choosing
VCAs to achieve a visually aesthetic result given an
input image. In this context, the workflow consists of
two primary steps: (1) the user has to select a VCA,
and (2) the user has to find proper parameter settings.
Moreover, these two steps can reoccur several times
forming a VCA pipeline.
Assisting the user in the choosing process poses
many challenges. Besides a recommendation system,
a manageable user interface must be implemented that
hides the overwhelming complexity of the rendering
techniques. To face the challenge in the user expe-
rience of image filtering apps, valuable insights com-
prising a featured iOS App, BeCasso (Pasewaldt et al.,
2016) are obtained (Klingbeil et al., 2017).
3 SYSTEM
Preliminaries & Assumptions. For modeling low-
level and high-level use cases for the proposed recom-
mendation system, we make the following assump-
tions. First, we avoid to transfer the input image to
the system due to the following reasons: (1) to be ap-
plicable in different, possibly world-wide scenarios,
with varying data protection and data privacy regula-
tions, (2) to account for possibly limited data through-
put between device and service, and (3) to avoid re-
quiring respective storage capacities. Instead, we rely
on representing the input image characteristics using
common low-level image feature descriptors that are
computed on-device prior to transmission. Secondly,
to enable the mapping of data transmission and re-
trieval to a user we assume a unique user token.
System Overview. Figure 2 outlines the interaction
between a user, a Visual Media Abstraction System
A Service-Based Preset Recommendation System for Image Stylization Applications
253

DB
User
Analysis
Operations
Input
Descriptors
VCAs
Result
Descriptors
Input Image Result Image
Figure 2: High-level outline of the framework for data stor-
age and data analysis in the domain of visual computing
applications (Visual Media Abstraction System (VMAS)
is colored in orange and Data Storage and Analysis Sys-
tem (DSAS) is colored in green and blue).
(VMAS), and the Data Storage and Analysis System
(DSAS), which are briefly described as follows:
VMAS: The VMAS computes the result image
based on the input image and VCA selection.
VCAs comprises all information needed to per-
form the computation including parameter values
and additional metadata.
DSAS: This paper focuses on the DSAS which con-
stitutes the framework for data profiling of image
and VCA metrics. The term “framework” refers
to a system that facilitates the development of ana-
lytics applications leveraging the frameworks Ap-
plication Programming Interface (API) for storing
and retrieving usage data.
Usage Data Acquisition and Storage. The user
wants to edit an uploaded input image. The VMAS
will provide a selection of VCAs and parameters to
stylize the image according to the user’s input. On
the basis of these VCAs, the VMAS computes the re-
sult image which is sent back to the user subsequently.
During this process, there are several points where
useful data for the later analysis is generated:
Input Image Descriptors: The actual input image
contains useful image descriptors. These descrip-
tors can be easily retrieved in case of general in-
formation descriptors, such as spatial resolution,
filename or image type. Moreover, there are spe-
cific domain information descriptors that contain
information about the image such as object recog-
nition data.
VCA Data: Further, VCA data including selected
parameters determine the computation of the re-
sulting image. This data is most important for the
later analysis as it tracks how the user interacts
with the given VCAs and parameters to produce a
certain result image.
Output Image Descriptors: The resulting image
usually comprises the same type of descriptors as
the input image. This information can be useful to
understand the user’s perception of visual aesthet-
ics. By comparing the input and output descrip-
tors, more subtle insights can be detected about
desired visual changes made by the user.
Recommendation System. The user wants a rec-
ommendation of suitable VCAs and parameter set-
tings to produce a visually aesthetic result image. The
recommendation is based on the analysis of the usage
data, which is stored in the DSAS.
4 IMPLEMENTATION ASPECTS
This section covers implementation details of the fun-
damental software framework used (Section 4.1) and
the database modeling (Section 4.2).
4.1 Software Framework
Figure 3 shows an overview of the proposed frame-
work that basically consists of two microservices: the
Usage Data Collector (UDC) and Usage Data An-
alyzer (UDA). Together these form the conceptual
DSAS. Both have access to the same database, which
holds the usage data at hand.
The framework has to provide an API to the raw
data and general statistic routes. To achieve better
(1) maintainability, (2) stability, and (3) separation
of concern the data service and analytics are estab-
lished as two separate services. Presuming there will
be more analytics use cases in the future, only the
UDA must be adapted, whereas the UDC remains
unaffected. Furthermore, this separation has perfor-
mance advantages in regard to handling incoming re-
quests. Thus, the UDA can be specialized in receiving
statistics requests and analyzing results. On the other
hand, the UDC handles primarily usage data import
requests. The general workflow of using this frame-
work is as follows:
1. An application for image processing sends an Hy-
pertext Transfer Protocol (HTTP) request contain-
ing image descriptors and a VCA pipeline config-
uration to the UDC.
2. The UDC validates the incoming usage data and
stores it in the GDMS.
3. An application sends an HTTP request to the
UDA in order to get recommendations, e.g., the
most used operations for a specific user.
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
254

Usage Data Collector
Send usage data
Request recommendations
GDMS
Application
(Mobile, Web)
Usage Data Analyzer
WRITE
READ
Figure 3: Overview of the services architecture and major
requests.
4. The UDA runs an analysis on the usage data
stored in the GDMS and sends the result as re-
sponse to the application.
Note that the UDA only has to read data from the
database, whereas the UDC must be able to insert
data. Since this paper involves the mere collection
and analysis of usage data, update and delete opera-
tions were omitted for the sake of brevity.
4.2 Database Modeling
This section describes the preliminaries and chal-
lenges for modeling the data base of the proposed rec-
ommendation system as well as implementation de-
tails specific to the individual services.
Preliminaries and System Challenges. Initially, a
recommendation system is challenged by the follow-
ing conceptual and technical aspects (Goyani and
Neha, 2020), which can be aligned to our problem
domain as follows:
Cold Start Problem: Recommendation techniques
that rely on existing data, e.g., based on user inter-
actions with an application, are not able to com-
pute recommendations for a new user as no usage
data about this user is collected yet. In this case,
recommendations must be made based on other
metrics, such as most used operations.
Data Sparsity: Some recommendation techniques
use matrices to store and compute scores for,
e.g., user-operation-relations, which indicate how
much an operation is liked by a user. If many
users interact with few operations only, this score
matrix is sparse and can consume storage space
unnecessarily.
Neo4j Dgraph TigerGraph JanusGraph
Open Source
✅ ✅ ❌ ✅
Self hostable
✅ ✅ ❌ ✅
Stable / Mature Ecosystem
✅ ✅ ✅ ❌
API for JavaScript
✅ ✅ ❌ ❌
Complex Graph / ML algorithms
✅ ❌ ✅ ❌
Figure 4: Feature comparison of different graph-based
databases with respect to their suitability for our approach.
Scalability: As computation performance decreases
with the amount of stored date, it should be able to
deploy more instances of the UDC and UDA mi-
croservices even on multiple machines in a cloud-
based environment.
Adaptability: There are several approaches for com-
puting recommendations (Das et al., 2017), yet we
have no data indicating which approaches fit best
for our use cases. Therefore, we want to be able to
implement several approaches and compare them
to each other, based on accuracy or performance.
In this work, we are only implementing recom-
mendations based on number or count of usage.
But as a requirement, all services should be adapt-
able to support other approaches as well.
Structural Flexibility: Since the system should be
expandable in terms of recommendation ap-
proaches, we require our data to have a flexible
structure. Having no rigid schema, allows us to
add or change properties or adapt the underlying
storage structure on demand.
Assessment of DBMS Approaches. Prior to
database implementation, we have to choose a
suitable Database Management System (DBMS)
approach. We assess the suitability between the
following three approaches:
Relational DB: This classical DB approach offers
only limited structural flexibility due to the usage
of a static database schema. Further, it requires
often expensive JOIN operations on large tables
to perform query operations.
Document-based DB (NoSQL): While document-
based approaches allow for structural flexibility
due to schema less model, it is difficult to express
and represent related data. All related data must
be stored in a document, which means there are
many duplicates which makes it hard to analyze
the data.
Graph-based DB (NoSQL): Graph-based ap-
proaches allow for structural flexibility due to
schema less model. Relations are represented as
edges in a graph, which means there are no JOIN
operations required to query related data. This
makes querying usually more performant than in
relational DBs, especially for big data amounts.
With respect to the latter approach, Figure 4 shows
comparison results of four different alternatives.
Database Schema. Figure 5 shows the database
schema developed for the prototypical implementa-
tion of our concept. It represents the following struc-
A Service-Based Preset Recommendation System for Image Stylization Applications
255

:App
name
platform
version
:User
id
:Medium
type: image
:Metadata
type: EXIF
focalLength: 4.25
colorSpace: 65535
:Metadata
type: JFIF
xDensity: 72
yDensity: 72
:Metadata
type: GPS
latitude: 52.382
longitude: 13.08
:Pipeline
preset:
winterMarcOil
:Operation
id: oil_0
preset
:Operation
id: styleTransfer
preset
:Parameter
name: details
value: 8.5
:Parameter
name: contrast
value: 1
Figure 5: Database schema depicting nodes and and rela-
tions for our recommendation system for image stylization
approaches.
ture on a per-node basis. A medium (image) relates to
an specific application that has uploaded the data and
a user that initiated the editing. Further, a medium
is described by to an number of metadata nodes, en-
coding standard attributes or descriptors. These rela-
tions can be extended by additional metadata format
for future extension. Furthermore and foremost, each
medium node relates with a pipeline node that pro-
cessed this medium. A pipeline references the indi-
vidual operations with its respective order of process-
ing and the respective preset used. Each operation
relates the to operation parameters that were manipu-
lated by the user and its respective value.
5 PRELIMINARY EVALUATION
This section describes and discusses the preliminary
evaluation of our approach by means of application
examples and performance measurements.
5.1 Application Example
In the following, the described microservice-based
framework is integrated into a Web application, which
is exemplary for any mobile or web application. The
application requests the most used operations from
the UDA to receive a sorted list of popular opera-
tions used. As shown in Figure 6, the list is used to
place popular operations more prominently by rank-
ing them before other operations and labeling them
with a star symbol.
If the user wants to adapt the style of the image
even further, they can use parameters or predefined
presets. Again, the application requests the most used
presets from the UDA to get a list of popular pre-
sets for the selected operation. Figure 6 shows how
most used presets are highlighted in the application
by adding the star symbols. In conclusion, a user can
select a featured VCA and preset using a few clicks.
(a) Sorting after featured VCAs.
(b) Selection of featured preset.
Figure 6: Integration of the prototypical implementation of
the recommendation system into a web application for im-
age stylization.
5.2 Performance Evaluation
As part of a feasibility study, we tested the perfor-
mance of the the UDC and UDA system components.
Test Data & System. For testings purposes we used
a data set comprising 11 609 entries. The complete
data set comprises 462190 nodes in the database, ap-
proximately 40 nodes per entry. The nodes are dis-
tributed as follows (Figure 5): 472 application nodes,
11594 medium respective pipeline nodes, 35 696
metadata nodes, 54345 operation nodes, and 348 488
parameter nodes. For testing purposes, we deployed
the UDC and UDA system components on a sin-
gle machine with the following system specifications:
Intel Core i9 10900K 3.70 GHz Central Processing
Unit (CPU), 10 (20 logical) cores, and 64 GB Ran-
dom Access Memory (RAM). The Node.js Version
18.6.0 runtime environment runs a Neo4j Version
4.4.7 database.
Performance Results. Figure 7 shows the resulting
measurements of our performance experiment: the
successive import of the 11 609 usage data entries into
an empty Database (DB). On a per-run basis, we
subsequently measured the runtime of (1) each data
import-query (Figure 7) and (2) querying the Top-10
operations among all applications nodes (Figure 7).
The test results show, that the duration of data import
queries are almost constant at approximately 60 ms.
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
256

0
20
40
60
80
100
120
140
160
180
200
1 502 1002 1503 2003 2503 3003 3503 4007 4507 5007 5507 6008 6508 7009 7509 8009 8509 9009 9510 10010 10512 11015 11515
Duration in Milliseconds
Request Number
(a) Import usage data performance
0
10
20
30
40
50
60
70
1 502 1002 1503 2003 2503 3003 3503 4007 4507 5007 5507 6008 6508 7009 7509 8009 8509 9009 9510 10010 10512 11015 11515
Duration in Milliseconds
Request Number
(b) Retrieval of most-used operations performance
Figure 7: Performance measurements for database input
and retrieval queries. Yellow lines show individuals tim-
ing while the line shows averaged values, respectively.
However, the runtime performance of retrieving the
most-used operations increase linearly with the num-
ber of stored database nodes.
5.3 Future Work
In our framework, we implemented a basic recom-
mendation system that evaluates and ranks operations
and presets based on the amount they are used. This
prototypical implementation can be extended by us-
ing more sophisticated state-of-the-art recommenda-
tion approaches, such as content-based, collaborative,
or hybrid filtering (Das et al., 2017). This includes
techniques that involve Machine Learning, e.g., for
finding clusters and relations between images and
VCAs and parameters that were used to stylize it.
In terms of scalability, the framework can be im-
proved by integrating a caching or pre-computing
workflow. Currently, the UDA analyzes the usage
data on incoming requests. This means, the run-
time performance of an analysis request increases
with the amount of stored usage data. To improve
the performance, the UDA could run the analysis
asynchronously, e.g., by periodically pre-compute the
most used operations and presets. The analysis results
can then be cached and served constantly on request.
This can be extended by using two database instances.
The first database is used by the UDC to store all the
usage data. Its performance is not degraded when an
analysis is running. The second database that holds
a copy of all the usage data, is able to mutate the
data, e.g., by adding labels used for similarity compu-
tations. The performance of analysis is not degraded
when new usage is collected at the same time. Addi-
tionally, no database transactions are required to syn-
chronize read and write operations to guarantee con-
sistency.
6 CONCLUSIONS
In conclusion, this paper presents a foundation for
all kinds of data profiling tasks, as well as VCA and
image data analytics. The presented framework con-
tains a usage data storing process enabled by a Us-
age Data Collector microservice and the underlying
Graph-based Database Management System. More-
over, the Usage Data Analyzer forms a basis for
data analysis tasks and already includes a prototypi-
cal analysis for retrieving the most popular operations
and presets. Finally, this work serves as an enabler for
various future use cases comprising the data storage
and analysis of image and VCA metrics.
ACKNOWLEDGMENTS
This work was partially funded by the Ger-
man Federal Ministry of Education and Research
(BMBF) through grants 01IS18092 (“mdViPro”) and
01IS19006 (“KI-LAB-ITSE”).
REFERENCES
Abedjan, Z., Golab, L., and Naumann, F. (2017). Data pro-
filing: A tutorial. In Proceedings of the 2017 ACM In-
ternational Conference on Management of Data, SIG-
MOD ’17, page 1747–1751, New York, NY, USA. As-
sociation for Computing Machinery.
Aciar, S. V., Zhang, D., Simoff, S. J., and Debenham, J. K.
(2006). Recommender system based on consumer
product reviews. 2006 IEEE/WIC/ACM International
Conference on Web Intelligence (WI 2006 Main Con-
ference Proceedings)(WI’06), pages 719–723.
Beri, B. and Singh, P. (2013). Web analytics: Increasing
website’s usability and conversion rate. International
Journal of Computer Applications, 72:35–38.
Das, D., Sahoo, L., and Datta, S. (2017). A survey on rec-
ommendation system. International Journal of Com-
puter Applications, 160:6–10.
D
¨
urschmid, T., S
¨
ochting, M., Semmo, A., Trapp, M., and
D
¨
ollner, J. (2017). ProsumerFX: Mobile Design of
Image Stylization Components. In Proceedings SIG-
GRAPH ASIA Mobile Graphics and Interactive Appli-
cations (MGIA), pages 1:1–1:8, New York. ACM.
Gatys, L. A., Ecker, A. S., and Bethge, M. (2016). Im-
age style transfer using convolutional neural networks.
A Service-Based Preset Recommendation System for Image Stylization Applications
257

2016 IEEE Conference on Computer Vision and Pat-
tern Recognition (CVPR), pages 2414–2423.
Goyani, M. and Neha, C. (2020). A review of movie rec-
ommendation system. ELCVIA: electronic letters on
computer vision and image analysis, 19(3):18–37.
Han, J., Kamber, M., and Pei, J. (2011). Data Mining: Con-
cepts and Techniques. Morgan Kaufmann Publishers
Inc., San Francisco, CA, USA, 3rd edition.
Huang, Z., Chung, W., Ong, T.-H., and Chen, H. (2002). A
graph-based recommender system for digital library.
In Proceedings of the 2nd ACM/IEEE-CS Joint Con-
ference on Digital Libraries, JCDL ’02, page 65–73,
New York, NY, USA. Association for Computing Ma-
chinery.
Isenberg, T. (2016). Interactive NPAR: What Type of Tools
Should We Create? In B
´
enard, P. and Winnem
¨
oller,
H., editors, Non-Photorealistic Animation and Ren-
dering. The Eurographics Association.
Klingbeil, M., Pasewaldt, S., Semmo, A., and D
¨
ollner, J.
(2017). Challenges in user experience design of im-
age filtering apps. In SIGGRAPH Asia 2017 Mobile
Graphics Interactive Applications, SA ’17, New York,
NY, USA. Association for Computing Machinery.
Kumar, M., Yadav, D., Singh, A. K., and Gupta, V. K.
(2015). A movie recommender system: Movrec. In-
ternational Journal of Computer Applications, 124:7–
11.
Moscato, V., Picariello, A., and Sperl
´
ı, G. (2021). An emo-
tional recommender system for music. IEEE Intelli-
gent Systems, 36:57–68.
Nouri, E., Sim, R., Fourney, A., and White, R. W. (2020).
Step-wise recommendation for complex task support.
In Proceedings of the 2020 Conference on Human In-
formation Interaction and Retrieval, CHIIR ’20, page
203–212, New York, NY, USA. Association for Com-
puting Machinery.
Pasewaldt, S., Semmo, A., D
¨
ollner, J., and Schlegel, F.
(2016). BeCasso: Artistic Image Processing and Edit-
ing on Mobile Devices. In Proceedings of ACM SIG-
GRAPH ASIA Mobile Graphics and Interactive Appli-
cations (MGIA), pages 14:1–14:1, New York. ACM.
Ricci, F., Rokach, L., and Shapira, B. (2022). Recommender
Systems: Techniques, Applications, and Challenges,
pages 1–35. Springer US, New York, NY.
Ricci, F., Rokach, L., Shapira, B., and Kantor, P. B. (2010).
Recommender Systems Handbook. Springer-Verlag,
Berlin, Heidelberg, 1st edition.
Richter, M., S
¨
ochting, M., Semmo, A., D
¨
ollner, J.,
and Trapp, M. (2018). Service-based Processing
and Provisioning of Image-Abstraction Techniques.
In Proceedings International Conference on Com-
puter Graphics, Visualization and Computer Vision
(WSCG), pages 97–106, Plzen, Czech Republic. Com-
puter Science Research Notes (CSRN).
Shadija, D., Rezai, M., and Hill, R. (2017). Microservices:
Granularity vs. performance. In Companion Proceed-
ings of The10th International Conference on Utility
and Cloud Computing, UCC ’17 Companion, page
215–220, New York, NY, USA. Association for Com-
puting Machinery.
Sperl
´
ı, G., Amato, F., Mercorio, F., Mezzanzanica, M.,
Moscato, V., and Picariello, A. (2018). A social me-
dia recommender system. Int. J. Multim. Data Eng.
Manag., 9:36–50.
Viggiato, M., Terra, R., Rocha, H., Valente, M. T., and
Figueiredo, E. (2018). Microservices in practice: A
survey study. CoRR, abs/1808.04836.
Xia, B. and Gong, P. (2014). Review of business intelli-
gence through data analysis. Benchmarking: An In-
ternational Journal, 21:300–311.
¨
Olveck
´
y, M. and Host’oveck
´
y, M. (2021). Digital image
forensics using exif data of digital evidence. In 2021
19th International Conference on Emerging eLearn-
ing Technologies and Applications (ICETA), pages
282–286.
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
258