Semantic Pattern-based Retrieval of Architectural Floor Plans with
Case-based and Graph-based Searching Techniques and their Evaluation
and Visualization
Qamer Uddin Sabri
1,2,?
, Johannes Bayer
1,2,?
, Viktor Ayzenshtadt
1,3,?
,
Syed Saqib Bukhari
1
, Klaus-Dieter Althoff
1,3
and Andreas Dengel
1,2
1
German Research Center for Artificial Intelligence, Trippstadter Strasse 122, 67663 Kaiserslautern, Germany
2
Technical University Kaiserslautern, P.O. Box 3049, 67663 Kaiserslautern, Germany
3
University of Hildesheim, Institute of Computer Science, Samelsonplatz 1, 31141 Hildesheim, Germany
Keywords:
Graph Matching, Subgraph Matching, Graph Isomorphism, Architectural Floor Plan, Case-based Reasoning,
Pattern Recognition.
Abstract:
Until today, for the conceptual design of architectural floor plans, architects widely follow the traditional pen
and paper based method to draw the conceptual floor plans, and retrieve the similar floor plans in the printed
reference collections. In this paper we present a complete end-to-end system that helps architects to retrieve
similar floor plans in early design phases. This work makes a three-fold contribution. Firstly, we have adapted
three state of the art techniques to retrieve the similar floor plans: case-based reasoning (CBR), exact graph
matching, and inexact graph matching. Secondly, we conducted a test to detect the computational limits of
the searching techniques. And finally, we performed a qualitative analysis by running more realistic test cases
created by architects while keeping in mind the computational limits. For visualization of results, we have
integrated advanced version of our previously implemented web-based user interface. The qualitative analysis
showed that the exact graph matching gives in general better results for a majority of test cases, as compared
to other two methods. The novelty of our approach is that it combines CBR, exact, and inexact graph matching
in one system in the domain of retrieval of architectural floor plans.
1 INTRODUCTION
When starting the design of a building, architects have
to develop floor plan concepts facing usually only
vague description of the building’s requirements. In
order to get inspiration to solve the creative problems,
working with references is an established method
in architecture. Traditionally, this leads to a labor-
intensive search and review of magazines and books
to find ideas similar to an initial concept. In order to
speed up this process, a dedicated search automation
would be needed.
For this purpose, we have already presented a
distributed case-based retrieval approach MetisCBR
(Ayzenshtadt et al., 2016a) for search of similar archi-
tectural building designs which is prototypically im-
plemented as part of the infrastructure of a basic re-
search project Metis – Knowledge-based search and
query methods for the development of semantic infor-
?
Equal contribution to the paper.
mation models (BIM) for use in early design phases.
Metis is an interdisciplinary project, funded by the
German Research Foundation (Deutsche Forschungs-
gemeinschaft, DFG).
MetisCBR can be accessed by a dedicated user in-
terface called WebUI (see Figure 1). This web-based
GUI allows for creating architectural concepts (floor
plans) and sending them to the retrieval engines. The
general usability of the WebUI has been shown by the
means of a user study (Bayer et al., 2015).
However, so far other types of well-known search
methods, like graph and subgraph matching-based re-
trieval, are not fully tested on this problem, as well as
not compared directly with CBR-based approaches.
In this paper we present a complete end-to-end
architectural design support system called Archis-
tant (http://www.dfki.uni-kl.de/archistant), that im-
plements the previously mentioned WebUI that lets
the user draw the required floor plan with rooms and
connection between these rooms, adjust the search
50
Sabri, Q., Bayer, J., Ayzenshtadt, V., Bukhari, S., Althoff, K-D. and Dengel, A.
Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation and Visualization.
DOI: 10.5220/0006112800500060
In Proceedings of the 6th Inter national Conference on Pattern Recognition Applications and Methods (ICPRAM 2017), pages 50-60
ISBN: 978-989-758-222-6
Copyright
c
 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

Figure 1: Screenshot of the Archistant WebUI. As a room oriented tool, the individual rooms may be dragged independently
from each other, revealing their editable connections to each other. Single lines represent wall connections, double lines
represent doors.
User Request
Merging of Results
WebUI
Us
er
Ra
ting
Log
MetisCBR
Index
Based
Retrieval
VF2
Retrieval
DatabaseRetrieval Systems
Result Processing
Client
Floorplans
Neo4j
Se
rv
e
r
Re
s
po
n
s
e
Processing
Layer
Figure 2: Overview over the system architecture of Archistant (simplified).
criteria, search the similar floor plans, and then visu-
alize the mapping between user’s search query (floor
plan) and resulted floor plans. For retrieval of sim-
ilar floor plans, we have implemented three search-
ing techniques, namely case-based retrieval (previ-
ously mentioned MetisCBR), exact graph matching,
and inexact graph matching. The processing pipeline
of Archistant is divided in three steps. First, the user
draws the search query with the WebUI. Afterwards
the query is forwarded to the search engines, each
of them extracts the semantic search patterns (finger-
prints) (see also Figures 2 and 3) and matches these
patterns with floor plans in the database. The result
sets of the systems are then unified and sent back to
the WebUI for visualization and room mapping.
We also conducted a boundary test of each re-
Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation
and Visualization
51

trieval method with respect to each search pattern to
analyze their technical limitations. Keeping in mind
these technical limitations, we performed a qualitative
analysis with some pre-defined search requests to de-
termine which retrieval method is suitable for which
user scenario.
2 RELATED WORK
This section provides an overview of work that is re-
lated to our paper. Related work can be divided into
three categories: CBR, graph and subgraph matching,
and sketch-based interfaces.
2.1 Case-based Reasoning
As mentioned in Section 1, MetisCBR uses methods
of CBR-based retrieval to find similar floor plans. A
comprehensive collections of work related to Metis-
CBR is described in detail in (Heylighen and Neuck-
ermans, 2001) and (Richter et al., 2007).
2.2 Graph and Subgraph Matching
Graph matching is being widely used these days for
the retrieval problems. It might be the case that there
is no exact match for the whole graph. To tackle this
situation another promising feature is provided that is
known as subgraph matching, that is, if two graphs
are not completely isomorphic but their part(s) are
isomorphic then they can be detected with subgraph
matching. An implementation that uses graph match-
ing to retrieve the similar floor plans is described in
(Ahmed et al., 2014). This work uses a modified
version of (Messmer and Bunke, 1999) for retrieval
of similar floor plans by arranging the row-column
vectors of the adjacency matrix in the decision tree.
A related work in (Wessel et al., 2008) implements
graph and subgraph isomorphism to check the simi-
larity between a query graph and building models in
the database, considering some constraints, that is, if a
query graph corresponds to the constraints, only then
it would be checked for similarity.
2.3 Sketch-based Interfaces
In the fields of engineering and architecture, different
implementations of the sketch-based interfaces exist
that let the user convey the idea by drawing it. A web-
based user interface WebUI has been created as part
of the previously mentioned research project Metis. It
lets the user draw the architectural floor plans elec-
tronically, set the different search criteria and, once
the results are retrieved, the user can interpret and vi-
sualize them with mapping between the query and re-
sults (i.e which part of the query relates to which part
of the result). For query construction, WebUI uses the
AGraphML (Langenhan, 2015) specification.
3 FLOOR PLANS RETRIEVAL
TECHNIQUES
To accomplish the task of retrieval of building designs
with similar context, we developed a unified retrieval
framework (as part of Archistant) with currently three
methods implemented. Besides MetisCBR, we have
added the integration of two graph-based retrieval
techniques: the exact matching method with the VF2
algorithm, and the inexact matching method with the
underlying index-based data structure. The main con-
cern we had during the development of the framework
was the matter of fact that in real-world scenarios the
data is not always consistent, i.e, in some cases the
numbers of nodes and/or edges are not equal in the
two graphs. This does not allow for determination of
the exact isomorphism between these graphs. To deal
with such cases, our algorithms should be able to tol-
erate these errors and instead of finding the only best
solution, they should try to find a set of appropriate
solutions. This knowledge led us to the implementa-
tion of the semantic fingerprint patterns concept in all
of the three retrieval methods.
This section is further organized as follows: Sec-
tion 3.1 defines the semantic fingerprint, i.e., the
search patterns concept, Section 3.2 defines the graph
structure, and then the descriptions of the three match-
ing techniques using these fingerprints are presented
in Section 3.3.
3.1 Semantic Fingerprints Concept
The concept of semantic fingerprint of architecture
(Langenhan and Petzold, 2010) is based on an index-
based hierarchical structure of building data. The hi-
erarchical nature of semantic fingerprints allows for
abstract representation of building topology and re-
lational connections between its parts. A fingerprint
pattern is a predefined prototype of such a structure
and is represented as a graph (see Figure 3) with op-
tions to have node labels (names of rooms) and edge
lables (names of edges/connections between rooms).
By using the idea of semantic fingerprint, our re-
trieval methods are able to either decompose (graph
and subgraph matching based techniques) the user
query into search patterns (fingerprints) or to derive
ICPRAM 2017 - 6th International Conference on Pattern Recognition Applications and Methods
52

the fingerprints from the query (MetisCBR), and then
match them with other floor plans in our database.
AGraphML is the XML-based representation for
floor plans. The specification (Langenhan, 2015) of
AGraphML defines attributes for this representation.
It can represent a complete architecture for a floor
plan, including the nodes and edges with their at-
tributes.
Figure 3 shows a list of 7 fingerprints currently
implemented in each of our retrieval engines.
3.2 Query Structure
Search queries and floor plans in our database are rep-
resented as graph G = (V, E). Here, V represents the
vertices (nodes) in the graphs, which are rooms in our
case, e.g., Kitchen or Sleeping. E represents the edges
(the connections between nodes), e.g., Kitchen and
Sleeping are connected with an edge called Door.
3.3 Matching Techniques
This section presents the three matching and retrieval
techniques implemented in Archistant: MetisCBR,
exact graph matching (referred as VF2), and inexact
graph matching (referred as index-based).
3.3.1 MetisCBR
MetisCBR is based on a decentralized structure where
agents and agent groups (retrieval containers) work on
different, predefined and assigned, tasks to support
an architect during the early design phases. The ac-
tual retrieval process is coordinated by a special agent
group (MetisCBR Coordinator) described in (Ayzen-
shtadt et al., 2016b). Being a case-based retrieval sys-
tem, MetisCBR’s theoretical emphasis is built on the
basic CBR paradigm that similar problems have sim-
ilar solutions. The system analyses the given prob-
lem (user’s search request) and tries to find solu-
tions (search results) that can be the most helpful to
solve this problem, that is, adapt the currently cre-
ated design according to findings in the search result.
MetisCBR is related to other systems and approaches
that were developed for the purpose of assisting an
architect, some examples are FABEL (Voss, 1997),
CBArch (Cavieres et al., 2011), or CaseBook (Inanc,
2000). To define a case (architectural design graph)
within the system, a specific distributed model (see
Figure 4) is applied to each graph-based floor plan.
The floor plans are then imported into the case base
of the system, where they are separated into three
main concepts: floor plan meta data, rooms and edges.
Each of the concepts consists of a number of specific
attributes (see also (Ayzenshtadt et al., 2015)). Dur-
ing the search, the weighting of single attributes and
an amalgamation of them is applied to find the most
similar concepts for the given search request (or its
parts).
The amalgamations are divided into fingerprint-
related and generic. The fingerprint-related amalga-
mations are applied only if one or more semantic
fingerprint patterns were added to the query. Each
of the fingerprint amalgamations considers only at-
tributes that were defined for this particular pattern.
This process is related to the footprint sets-based re-
trieval described in (Smyth and McKenna, 1999). In
contrast to the fingerprint-related, the generic amalga-
mation functions use all of the attributes, and are ap-
plied if no fingerprint definitions were found. Both of
the amalgamation types can be executed within differ-
ent retrieval strategies. Currently, two kinds of strate-
gies are implemented in the system: a multistep re-
sults preselection-based strategy for complex finger-
print patterns as well as for the deep search without
fingerprints (described in (Ayzenshtadt et al., 2015)),
and a single-step strategy for general retrieval with or
without fingerprints . A post-retrieval weighting pro-
cess (available only for search requests with multiple
fingerprints) can be applied to boost results for fin-
gerprints that were considered more important by the
user. The results of the retrieval are returned sorted in
descending order by the computed confidence score.
3.3.2 Exact Graph Matching Method (VF2)
In exact graph matching, one-to-one mapping is
known as isomorphism. When two graphs contain the
same number of nodes, and they are connected in the
same way, then we can call them isomorphic. Isor-
morphism can be found with exact graph matching
(Bengoetxea, 2002). For example, (Ullmann, 1976),
(Schmidt and Druffel, 1976), and (McKay et al.,
1981) are one-to-one exact graph-based matching ap-
proaches. Our system (Archistant) uses implemen-
tation of the VF2 algorithm, proposed in (Cordella
et al., 2004) (and implemented in the NetworkX li-
brary), because it significantly reduces the memory
requirement and obtains the best performance for
graphs of small size and for quite sparse graphs (Fog-
gia et al., 2001).
In our system, VF2-based method uses
AGraphML files to generate the graphs. Firstly,
AGraphML files from Neo4j database are generated,
as an offline step, with a tool named ”Neo4j Shell
Tools”. One AGraphML file will be generated against
each floor plan. Now our system can match the search
request with other floor plans of Neo4j database.
The step by step details of how VF2 system
Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation
and Visualization
53

Fingerprint Name Description Specics
FP1 Room Count Number of rooms No connections between
rooms and no labels
specied
FP2 Relation Count Number of edges No room information
specied
FP3 Room Graph Anonymous
representation of rooms
and edges
No labels specied for
rooms and edges
FP4 Room Types Labels of rooms No connections between
rooms only room labels
are specied
FP5 Adjacency Emphasis on room
semantics
Rooms information is
complete no edge labels
specied
FP6 Accessibility Emphasis on edge
semantics
Edge information is
complete no room labels
specied
FP7 Full Graph Complete graph All information about
rooms and edges
available
Figure 3: Fingerprints (i.e., search patterns) allow for abstract representation of building topology and relational connections
between its parts. Fingerprints currently implemented in all examined retrieval methods are shown here. Nodes represent the
rooms, edges represent the room connections.
matches (see Figure 5) search request with floor plans
is described as follows: in the first step, the system re-
ceives the search request, then it checks for its valid-
ity, the system proceeds if the request is valid. Then,
from the search request, AGraphML is extracted to
generate a graph, referred as query-graph, and sub-
sequently its fingerprints are generated. After having
fingerprints for query-graph, VF2 system then one by
one takes each of the offline generated AGraphML
files, generates its graph, referred as db-graph, and
its fingerprints, and then matches the correspond-
ing fingerprints, i.e., FP1 of the query-graph will be
matched against FP1 of db-graph of each AGraphML,
FP2 of query-graph with FP2 of db-graph of each
AGraphML and so on. Once the system is done
with matching fingerprints, it delivers the resulting
floor plans with confidence score in descending or-
der, the confidence score shows how closely a floor
plan matches with the user’s search request according
to each of the fingerprints in Figure 3.
3.3.3 Inexact Graph Matching Method
(Index-based)
There are different types of inexact, index-based
graph searching methods: GraphGrep (Giugno and
Shasha, 2002), Lucene index (Sharanya Jayaraman,
2013), FG-Index (Cheng et al., 2007), cIndex (Chen
et al., 2007). In this paper, we are using the Lucene
index-based method as it comes with the Neo4j
framework by default.
The index-based method operates as follows (see
Figure 6): A search request is decomposed into the
different fingerprints. The fingerprints are represented
as Cypher (Neo4j query language) queries and the
ones whose weight is different from 0 are transmit-
ted to the Neo4j server. The Neo4j server answers
the requests with a set of floor plans for each finger-
ICPRAM 2017 - 6th International Conference on Pattern Recognition Applications and Methods
54

FLOOR
PLAN
ROOM EDGE
is part-of
is part-of is part-of
Floorplan ID
Figure 4: The general description of the underlying distributed model for case-based retrieval in MetisCBR.
Neo4j
Search
Request
Graph
Graph
Final
Result List
O
f
f
l
i
n
e
G
e
n
e
r
a
t
i
o
n
Fingerprint
generation
AGraphML
AGraphML
VF2 Exact Matching System
Fingerprint
generation
Fingerprints
Matching
Figure 5: Above is the flow diagram of the VF2 exact matching system. It shows that from the search request first the
AGraphML, then the graph and finally the fingerprints are generated. In the next step, the search request’s fingerprints are
matched with the corresponding fingerprints of the offline generated AGraphML files. Finally, the results are transferred to
the requester.
print. These sets are unified and the floor plan results
are ordered according to the similarity score which is
calculated for every item in the final result set. This
similarity score is the sum of the user-defined weights
of the fingerprints for which the query matches the
database entry.
A fingerprint is considered to match if the fin-
gerprints graph is the subgraph of the database en-
try. The fingerprints are processed independently for
simplicity reasons, hence one room in the query may
be mapped to different rooms within the same floor
plan in the database. Figure 7 illustrates an exam-
ple of the fingerprints processing within the index-
based method. The query consists of three rooms la-
beled as Living, Kitchen and Sleeping. The Living
room is connected with Kitchen via an edge connec-
tion labeled as Passage, the Kitchen is connected with
Sleeping via an edge connection labeled as Wall, and
Sleeping room is connected with Living via an edge
connection labeled as Door. The right side of the di-
agram shows exemplary matching and non matching
fingerprints between search query and floor plan in
the database.
4 EVALUATIONS OF OUR
SYSTEM
4.1 Technical Limitations (Boundary
Test)
Because of the computational demands of matching
and other problems, all retrieval methods have their
Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation
and Visualization
55

Figure 6: The index-based method flow diagram. A search request is decomposed into AGraphML and fingerprint weights.
From the AGraphML the fingerprints are created and the fingerprints are translated into Cypher queries. The Cypher queries
are send to the database, each Cypher-translated fingerprint is answered with a set of floor plans. The result sets are unified.
Each floor plan is scored by the sum of the weights of the fingerprints corresponding to the the result sets that contain the floor
plan. Finally, the result set is sorted by the score and transferred to the requester.
own limitations regarding the complexity of search re-
quests they can handle, especially with respect to each
of the fingerprints shown in Figure 3. Given enough
time and storage, all search queries should be answer-
able in theory, but since computation time must be
limited due to usability reasons and storage is limited
due to economical reasons, the question arises how
complex the search requests may be so that they can
be reasonably processed by a given system. In order
to test our retrieval methods, we employed the follow-
ing method: for each type of fingerprint we generated
test scenarios of variable complexity which we instan-
tiated over an interval between 1 and 25 units of com-
plexity, where 25 is selected empirically as a maxi-
mum limit. The metrics for these units differ for the
different fingerprints, but are either equal to the num-
ber of rooms in a test scenario or the number of edges
in a test scenario. The test scenarios are designed so
that the systems are advised to only take the finger-
print into account the scenario was designed for. The
most complex test scenario that could be handled by
the system is considered to be the system’s boundary
for that fingerprint. The results are shown in Figure
8 that contains the boundaries of the fingerprints we
tested on our systems. Both the VF2 exact matching
and MetisCBR executed all test scenarios flawlessly,
only the index-based system has limitations over the
maximum size of fingerprints FP3, FP5, FP6 and FP7.
These limitations came from internal timeout errors
of the underlying Neo4j database, most likely due to
a congestion of the database system. Knowing the
boundaries of our different methods for a given hard-
ware and time limitation, we defined test scenarios for
a subsequent, qualitative analysis (see Section 4.2) in
a more safe manner.
4.2 Qualitative Analysis
The evaluation of quality and helpfulness of the re-
trieval results of the above described retrieval meth-
ods was the main task of the subsequent special study,
where the mixed team of architects and computer
scientists created queries and rated collaboratively a
number of graphical and graph-based result represen-
tations of the search result sets. The confidence scores
(range R = 0.0..1.0), computed by the retrieval meth-
ods during the retrieval, were taken into account as
well.
As search queries, 10 AGraphML requests with
floor plan constructions of different complexity, were
used. Each of the presented retrieval methods was
queried with each of those AGraphML-based floor
plans and delivered accumulated result sets, sorted in
descending order by confidence score of the single re-
sult. All of the fingerprint patterns were applied to the
given queries. For this qualitative analysis we asked
different participants to observe the results of each of
the three retrieval methods, corresponding to each of
the 10 queries. Each participant rated the three meth-
ods on the scale of first, second and third or equal. In
Table 1 the results of the subjective qualitative analy-
sis from the participants of the study are shown. This
ICPRAM 2017 - 6th International Conference on Pattern Recognition Applications and Methods
56

Figure 7: The above diagram shows how the fingerprints are integrated in the retrieval process. The right side of the diagram
indicates which fingerprints of the search query are matched with which fingerprints of the floor plan in the data base. Cross
indicates no match while green sign indicates a match. The diagram illustrates this for index based method, which also
considers the inexact matching.
Figure 8: Boundary Test Results. Depicted are the results of the boundary test for each search technique and fingerprint.
More precisely, for every combination of search technique and fingerprint the diagram shows the complexity rating of the
most complex test scenario the system could handle successfully given a machine with certain computational and memory
limits. The metric of complexity differs for the test scenarios of different fingerprints.
table accumulates the overall rankings and presents
the summarized results for each query. The ranking
results are divided into two categories, where the first
category represents queries for which the correspond-
ing retrieval method was determined as the clear win-
ner, that is, was considered best by all participants.
The second category is similar to the first one, but
shows which method won the majority of rankings for
the particular query. In case of equality of votes, the
corresponding query is placed in the third column. In
the last column the percentage of queries dominated
by the particular method is given.
Table 1 shows that each of the three retrieval meth-
ods is good for some query cases. In Figure 9, we
show the two best results of three selected queries for
each search method, where for each of the three se-
Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation
and Visualization
57

Table 1: Accumulated results of the result sets for the selected queries. The first category represents queries for which the
corresponding retrieval method was considered best by all participants. The second category shows which method won the
majority of rankings for the particular query. In case of equality of votes, the corresponding query is placed in the third
column. The last column contains the percentage of queries dominated by the particular method.
Retrieval Method Queries won Queries won by majority Co-winner in Summarized results
VF2 Q3, Q4 Q6, Q7 Q8, Q10 50%
Index-Based – Q5 Q10 15%
MetisCBR Q2, Q9 Q1 Q8 35%
Figure 9: Computed similarity values of the result sets for the selected queries. Color codes represent the room purposes, the
first column contains the queries with rooms and assigned purposes. For each retrieval method the first two best single results
for the corresponding query are represented. Each single result box consists of a confidence score, the corresponding graph
representation with color codes, and the graphical floor plan representation. The result set considered best by the participants
is enclosed in a colorized box.
lected queries the best case of each of three retrieval
methods is represented, respectively, according to Ta-
ble 1. For better understanding, how the query and
the results looked like and which representations the
architects would use while working with our system,
we visualized them using both graph-based (the top
representation in Table 1) and graphical (the bottom
representation in Table 1) representations. This figure
does not include the subjective rating of the results,
only the confidence scores computed by the corre-
sponding retrieval method are presented. It is notable
that the CBR method has a higher score in relation
to other two methods. This can be explained with
the fact that the application of some fingerprints (e.g.,
Room Purposes, Room Count or Edge Count) caused
the finding of the (almost) exact matches of the room
label sets or number of the contained rooms or room
connections. This increased their confidence in the
accumulated result set. This difference in scoring is
observable in other result sets as well.
All things considered, the retrieval methods were
able to present sets of reasonable results, taking the
ICPRAM 2017 - 6th International Conference on Pattern Recognition Applications and Methods
58

current stage of the respective method development
into account. According to the subjective rankings
of results and in contrast to the computed confi-
dence score, the VF2 exact matching method was
the best one in the overall rating. This fact sup-
ports the assumption that exact matching is a better
method for determination of isomorphism in graph-
based datasets, where a set of tolerable technical lim-
itations is given.
5 CONCLUSION AND FUTURE
WORK
In this paper, we introduced Archistant, a complete
end-to-end system for search and retrieval of sim-
ilar floor plans for architects. By using this sys-
tem, we now have a complete automated pipeline
for sketching architectural concepts and searching for
similar ideas. We showed the capabilities of the im-
plemented retrieval techniques by means of both a
boundary test study (in order to show fundamental
performance capabilities and limitations of the meth-
ods) and a qualitative study (in order to show that the
methods produce reasonable results). In the boundary
test study, the VF2-based and MetisCBR replied flaw-
lessly and the index-based method at least answered
most queries without errors. In the qualitative anal-
ysis, each method showed at least in some situation
good performances. The currently implemented fin-
gerprints mainly take graph-based floor-ground prop-
erties into account. Geometric properties are not con-
sidered. But these information have a big potential for
making search results more accurate. A large-scale
user study resulting in a big ground-truth database
may be conducted in future and may hereby lead to
the deployment of machine learning for automatically
selecting the best approach for every search scenario
in future.
REFERENCES
Ahmed, S., Weber, M., Liwicki, M., Langenhan, C., Den-
gel, A., and Petzold, F. (2014). Automatic analysis
and sketch-based retrieval of architectural floor plans.
Pattern Recognition Letters, 35:91–100.
Ayzenshtadt, V., Langenhan, C., Bukhari, S. S., Althoff,
K.-D., Petzold, F., and Dengel, A. (2015). Dis-
tributed domain model for the case-based retrieval
of architectural building designs. In Petridis, M.,
Roth-Berghofer, T., and Wiratunga, N., editors, Pro-
ceedings of the 20th UK Workshop on Case-Based
Reasoning. UK Workshop on Case-Based Reason-
ing (UKCBR-2015), located at SGAI International
Conference on Artificial Intelligence, December 15-
17, Cambridge, United Kingdom. School of Com-
puting, Engineering and Mathematics, University of
Brighton, UK.
Ayzenshtadt, V., Langenhan, C., Bukhari, S. S., Althoff, K.-
D., Petzold, F., and Dengel, A. (2016a). Thinking with
containers: A multi-agent retrieval approach for the
case-based semantic search of architectural designs.
In Filipe, J. and van den Herik, J., editors, Proceed-
ings of the 8th International Conference on Agents
and Artificial Intelligence. International Conference
on Agents and Artificial Intelligence (ICAART-2016),
February 24-26, Rome, Italy. SCITEPRESS.
Ayzenshtadt, V., Langenhan, C., Roth, J., Bukhari, S. S.,
Althoff, K.-D., Petzold, F., and Dengel, A. (2016b).
Comparative evaluation of rule-based and case-based
retrieval coordination for search of architectural build-
ing designs. In Goel, A., Roth-Berghofer, T., and
Diaz-Agudo, B., editors, Case-based Reasoning in
Research and Development. International Conference
on Case-Based Reasoning (ICCBR-16), 24th Interna-
tional Conference on Case Based Reasoning, October
31 - November 2, Atlanta,, Georgia, USA. Springer,
Berlin, Heidelberg.
Bayer, J., Bukhari, S. S., Langenhan, C., Eichenberger-
Liwicki, M., Althoff, K.-D., Petzold, F., and Den-
gel, A. (2015). Migrating the classical pen-and-paper
based conceptual sketching of architecture plans to-
wards computer tools - prototype design and evalua-
tion.
Bengoetxea, E. (2002). Inexact graph matching using es-
timation of distribution algorithms. Ecole Nationale
Sup
´
erieure des T
´
el
´
ecommunications, Paris, 2:4.
Cavieres, A., Bhatia, U., Joshi, P., Zhao, F., and Ram, A.
(2011). CBArch: A Case-Based Reasoning Frame-
work for Conceptual Design of Commercial Build-
ings. Artificial Intelligence and Sustainable Design
– Papers from the AAAI 2011 Spring Symposium (SS-
11-02), pages 19–25.
Chen, C., Yan, X., Yu, P. S., Han, J., Zhang, D.-Q., and
Gu, X. (2007). Towards graph containment search
and indexing. In Proceedings of the 33rd international
conference on Very large data bases, pages 926–937.
VLDB Endowment.
Cheng, J., Ke, Y., Ng, W., and Lu, A. (2007). Fg-
index: towards verification-free query processing on
graph databases. In Proceedings of the 2007 ACM
SIGMOD international conference on Management of
data, pages 857–872. ACM.
Cordella, L. P., Foggia, P., Sansone, C., and Vento, M.
(2004). A (sub) graph isomorphism algorithm for
matching large graphs. Pattern Analysis and Machine
Intelligence, IEEE Transactions on, 26(10):1367–
1372.
Foggia, P., Sansone, C., and Vento, M. (2001). A perfor-
mance comparison of five algorithms for graph iso-
morphism. In Proceedings of the 3rd IAPR TC-15
Workshop on Graph-based Representations in Pattern
Recognition, pages 188–199.
Giugno, R. and Shasha, D. (2002). Graphgrep: A fast
and universal method for querying graphs. In Pattern
Semantic Pattern-based Retrieval of Architectural Floor Plans with Case-based and Graph-based Searching Techniques and their Evaluation
and Visualization
59

Recognition, 2002. Proceedings. 16th International
Conference on, volume 2, pages 112–115. IEEE.
Heylighen, A. and Neuckermans, H. (2001). A case base of
case-based design tools for architecture. Computer-
Aided Design, 33(14):1111–1122.
Inanc, B. S. (2000). Casebook. an information retrieval sys-
tem for housing floor plans. In The Proceedings of 5th
Conference on Computer Aided Architectural Design
Research (CAADRIA), pages 389–398.
Langenhan, C. (2015). A federated information system for
the support of topological bim-based approaches. Fo-
rum Bauinformatik Aachen.
Langenhan, C. and Petzold, F. (2010). The fingerprint
of architecture-sketch-based design methods for re-
searching building layouts through the semantic fin-
gerprinting of floor plans. International electronic
scientific-educational journal: Architecture and Mod-
ern Information Technologies, 4:13.
McKay, B. D. et al. (1981). Practical graph isomorphism.
Department of Computer Science, Vanderbilt Univer-
sity Tennessee, US.
Messmer, B. T. and Bunke, H. (1999). A decision tree ap-
proach to graph and subgraph isomorphism detection.
Pattern recognition, 32(12):1979–1998.
Richter, K., Heylighen, A., and Donath, D. (2007). Looking
back to the future-an updated case base of case-based
design tools for architecture. Knowledge Modelling-
eCAADe, 25:285–292.
Schmidt, D. C. and Druffel, L. E. (1976). A fast backtrack-
ing algorithm to test directed graphs for isomorphism
using distance matrices. Journal of the ACM (JACM),
23(3):433–445.
Sharanya Jayaraman, S. V. (2013). Comparative survey of
query processing on graph databases. Project report,
Florida State University.
Smyth, B. and McKenna, E. (1999). Footprint-based re-
trieval. In Case-Based Reasoning Research and De-
velopment, pages 343–357. Springer.
Ullmann, J. R. (1976). An algorithm for subgraph isomor-
phism. Journal of the ACM (JACM), 23(1):31–42.
Voss, A. (1997). Case design specialists in FABEL. Issues
and Applications of Case-based Reasoning in Design,
pages 301–335.
Wessel, R., Bl
¨
umel, I., and Klein, R. (2008). The room
connectivity graph: Shape retrieval in the architectural
domain.
ICPRAM 2017 - 6th International Conference on Pattern Recognition Applications and Methods
60