Refine
Year of publication
- 2015 (100) (remove)
Document Type
- Doctoral Thesis (100) (remove)
Has Fulltext
- yes (100)
Keywords
- Stadtplanung (2)
- finite element method (2)
- isogeometric analysis (2)
- tractor (2)
- verification (2)
- AMC225xe (1)
- Adaptive time step (1)
- Adjoint method (1)
- Allylierung (1)
- Aminierung (1)
Faculty / Organisational entity
- Kaiserslautern - Fachbereich Informatik (19)
- Kaiserslautern - Fachbereich Mathematik (19)
- Kaiserslautern - Fachbereich Maschinenbau und Verfahrenstechnik (18)
- Kaiserslautern - Fachbereich Chemie (17)
- Kaiserslautern - Fachbereich Elektrotechnik und Informationstechnik (10)
- Kaiserslautern - Fachbereich Biologie (6)
- Kaiserslautern - Fachbereich Raum- und Umweltplanung (4)
- Fraunhofer (ITWM) (3)
- Kaiserslautern - Fachbereich Bauingenieurwesen (3)
- Kaiserslautern - Fachbereich Sozialwissenschaften (2)
- Kaiserslautern - Fachbereich ARUBI (1)
The goal of this work is to develop statistical natural language models and processing techniques
based on Recurrent Neural Networks (RNN), especially the recently introduced Long Short-
Term Memory (LSTM). Due to their adapting and predicting abilities, these methods are more
robust, and easier to train than traditional methods, i.e., words list and rule-based models. They
improve the output of recognition systems and make them more accessible to users for browsing
and reading. These techniques are required, especially for historical books which might take
years of effort and huge costs to manually transcribe them.
The contributions of this thesis are several new methods which have high-performance computing and accuracy. First, an error model for improving recognition results is designed. As
a second contribution, a hyphenation model for difficult transcription for alignment purposes
is suggested. Third, a dehyphenation model is used to classify the hyphens in noisy transcription. The fourth contribution is using LSTM networks for normalizing historical orthography.
A size normalization alignment is implemented to equal the size of strings, before the training
phase. Using the LSTM networks as a language model to improve the recognition results is
the fifth contribution. Finally, the sixth contribution is a combination of Weighted Finite-State
Transducers (WFSTs), and LSTM applied on multiple recognition systems. These contributions
will be elaborated in more detail.
Context-dependent confusion rules is a new technique to build an error model for Optical
Character Recognition (OCR) corrections. The rules are extracted from the OCR confusions
which appear in the recognition outputs and are translated into edit operations, e.g., insertions,
deletions, and substitutions using the Levenshtein edit distance algorithm. The edit operations
are extracted in a form of rules with respect to the context of the incorrect string to build an
error model using WFSTs. The context-dependent rules assist the language model to find the
best candidate corrections. They avoid the calculations that occur in searching the language
model and they also make the language model able to correct incorrect words by using context-
dependent confusion rules. The context-dependent error model is applied on the university of
Washington (UWIII) dataset and the Nastaleeq script in Urdu dataset. It improves the OCR
results from an error rate of 1.14% to an error rate of 0.68%. It performs better than the
state-of-the-art single rule-based which returns an error rate of 1.0%.
This thesis describes a new, simple, fast, and accurate system for generating correspondences
between real scanned historical books and their transcriptions. The alignment has many challenges, first, the transcriptions might have different modifications, and layout variations than the
original book. Second, the recognition of the historical books have misrecognition, and segmentation errors, which make the alignment more difficult especially the line breaks, and pages will
not have the same correspondences. Adapted WFSTs are designed to represent the transcription. The WFSTs process Fraktur ligatures and adapt the transcription with a hyphenations
model that allows the alignment with respect to the varieties of the hyphenated words in the line
breaks of the OCR documents. In this work, several approaches are implemented to be used for
the alignment such as: text-segments, page-wise, and book-wise approaches. The approaches
are evaluated on German calligraphic (Fraktur) script historical documents dataset from “Wan-
derungen durch die Mark Brandenburg” volumes (1862-1889). The text-segmentation approach
returns an error rate of 2.33% without using a hyphenation model and an error rate of 2.0%
using a hyphenation model. Dehyphenation methods are presented to remove the hyphen from
the transcription. They provide the transcription in a readable and reflowable format to be used
for alignment purposes. We consider the task as classification problem and classify the hyphens
from the given patterns as hyphens for line breaks, combined words, or noise. The methods are
applied on clean and noisy transcription for different languages. The Decision Trees classifier
returns better performance on UWIII dataset and returns an accuracy of 98%. It returns 97%
on Fraktur script.
A new method for normalizing historical OCRed text using LSTM is implemented for different texts, ranging from Early New High German 14th - 16th centuries to modern forms in New
High German applied on the Luther bible. It performed better than the rule-based word-list
approaches. It provides a transcription for various purposes such as part-of-speech tagging and
n-grams. Also two new techniques are presented for aligning the OCR results and normalize the
size by using adding Character-Epsilons or Appending-Epsilons. They allow deletion and insertion in the appropriate position in the string. In normalizing historical wordforms to modern
wordforms, the accuracy of LSTM on seen data is around 94%, while the state-of-the-art combined rule-based method returns 93%. On unseen data, LSTM returns 88% and the combined
rule-based method returns 76%. In normalizing modern wordforms to historical wordforms, the
LSTM delivers the best performance and returns 93.4% on seen data and 89.17% on unknown
data.
In this thesis, a deep investigation has been done on constructing high-performance language
modeling for improving the recognition systems. A new method to construct a language model
using LSTM is designed to correct OCR results. The method is applied on UWIII and Urdu
script. The LSTM approach outperforms the state-of-the-art, especially for unseen tokens
during training. On the UWIII dataset, the LSTM returns reduction in OCR error rates from
1.14% to 0.48%. On the Nastaleeq script in Urdu dataset, the LSTM reduces the error rate
from 6.9% to 1.58%.
Finally, the integration of multiple recognition outputs can give higher performance than a
single recognition system. Therefore, a new method for combining the results of OCR systems is
explored using WFSTs and LSTM. It uses multiple OCR outputs and votes for the best output
to improve the OCR results. It performs better than the ISRI tool, Pairwise of Multiple Sequence and it helps to improve the OCR results. The purpose is to provide correct transcription
so that it can be used for digitizing books, linguistics purposes, N-grams, and part-of-speech
tagging. The method consists of two alignment steps. First, two recognition systems are aligned
using WFSTs. The transducers are designed to be more flexible and compatible with the different symbols in line and page breaks to avoid the segmentation and misrecognition errors.
The LSTM model then is used to vote the best candidate correction of the two systems and
improve the incorrect tokens which are produced during the first alignment. The approaches
are evaluated on OCRs output from the English UWIII and historical German Fraktur dataset
which are obtained from state-of-the-art OCR systems. The Experiments show that the error
rate of ISRI-Voting is 1.45%, the error rate of the Pairwise of Multiple Sequence is 1.32%, the
error rate of the Line-to-Page alignment is 1.26% and the error rate of the LSTM approach has
the best performance with 0.40%.
The purpose of this thesis is to contribute methods providing correct transcriptions corresponding to the original book. This is considered to be the first step towards an accurate and
more effective use of the documents in digital libraries.
Information Visualization (InfoVis) and Human-Computer Interaction (HCI) have strong ties with each other. Visualization supports the human cognitive system by providing interactive and meaningful images of the underlying data. On the other side, the HCI domain cares about the usability of the designed visualization from the human perspectives. Thus, designing a visualization system requires considering many factors in order to achieve the desired functionality and the system usability. Achieving these goals will help these people in understanding the inside behavior of complex data sets in less time.
Graphs are widely used data structures to represent the relations between the data elements in complex applications. Due to the diversity of this data type, graphs have been applied in numerous information visualization applications (e.g., state transition diagrams, social networks, etc.). Therefore, many graph layout algorithms have been proposed in the literature to help in visualizing this rich data type. Some of these algorithms are used to visualize large graphs, while others handle the medium sized graphs. Regardless of the graph size, the resulting layout should be understandable from the users’ perspective and at the same time it should fulfill a list of aesthetic criteria to increase the representation readability. Respecting these two principles leads to produce a resulting graph visualization that helps the users in understanding and exploring the complex behavior of critical systems.
In this thesis, we utilize the graph visualization techniques in modeling the structural and behavioral aspects of embedded systems. Furthermore, we focus on evaluating the resulting representations from the users’ perspectives.
The core contribution of this thesis is a framework, called ESSAVis (Embedded Systems Safety Aspect Visualizer). This framework visualizes not only some of the safety aspects (e.g. CFT models) of embedded systems, but also helps the engineers and experts in analyzing the system safety critical situations. For this, the framework provides a 2Dplus3D environment in which the 2D represents the graph representation of the abstract data about the safety aspects of the underlying embedded system while the 3D represents the underlying system 3D model. Both views are integrated smoothly together in the 3D world fashion. In order to check the effectiveness and feasibility of the framework and its sub-components, we conducted many studies with real end users as well as with general users. Results of the main study that targeted the overall ESSAVis framework show high acceptance ratio and higher accuracy with better performance using the provided visual support of the framework.
The ESSAVis framework has been designed to be compatible with different 3D technologies. This enabled us to use the 3D stereoscopic depth of such technologies to encode nodes attributes in node-link diagrams. In this regard, we conducted an evaluation study to measure the usability of the stereoscopic depth cue approach, called the stereoscopic highlighting technique, against other selected visual cues (i.e., color, shape, and sizes). Based on the results, the thesis proposes the Reflection Layer extension to the stereoscopic highlighting technique, which was also evaluated from the users’ perspectives. Additionally, we present a new technique, called ExpanD (Expand in Depth), that utilizes the depth cue to show the structural relations between different levels of details in node-link diagrams. Results of this part opens a promising direction of the research in which visualization designers can get benefits from the richness of the 3D technologies in visualizing abstract data in the information visualization domain.
Finally, this thesis proposes the application of the ESSAVis frame- work as a visual tool in the educational training process of engineers for understanding the complex concepts. In this regard, we conducted an evaluation study with computer engineering students in which we used the visual representations produced by ESSAVis to teach the principle of the fault detection and the failure scenarios in embedded systems. Our work opens the directions to investigate many challenges about the design of visualization for educational purposes.
In DS-CDMA, spreading sequences are allocated to users to separate different
links namely, the base-station to user in the downlink or the user to base station in the uplink. These sequences are designed for optimum periodic correlation properties. Sequences with good periodic auto-correlation properties help in frame synchronisation at the receiver while sequences with good periodic cross-
correlation property reduce cross-talk among users and hence reduce the interference among them. In addition, they are designed to have reduced implementation complexity so that they are easy to generate. In current systems, spreading sequences are allocated to users irrespective of their channel condition. In this thesis,
the method of allocating spreading sequences based on users’ channel condition
is investigated in order to improve the performance of the downlink. Different
methods of dynamically allocating the sequences are investigated including; optimum allocation through a simulation model, fast sub-optimum allocation through
a mathematical model, and a proof-of-concept model using real-world channel
measurements. Each model is evaluated to validate, improvements in the gain
achieved per link, computational complexity of the allocation scheme, and its impact on the capacity of the network.
In cryptography, secret keys are used to ensure confidentiality of communication between the legitimate nodes of a network. In a wireless ad-hoc network, the
broadcast nature of the channel necessitates robust key management systems for
secure functioning of the network. Physical layer security is a novel method of
profitably utilising the random and reciprocal variations of the wireless channel to
extract secret key. By measuring the characteristics of the wireless channel within
its coherence time, reciprocal variations of the channel can be observed between
a pair of nodes. Using these reciprocal characteristics of
common shared secret key is extracted between a pair of the nodes. The process
of key extraction consists of four steps namely; channel measurement, quantisation, information reconciliation, and privacy amplification. The reciprocal channel
variations are measured and quantised to obtain a preliminary key of vector bits (0; 1). Due to errors in measurement, quantisation, and additive Gaussian noise,
disagreement in the bits of preliminary keys exists. These errors are corrected
by using, error detection and correction methods to obtain a synchronised key at
both the nodes. Further, by the method of secure hashing, the entropy of the key
is enhanced in the privacy amplification stage. The efficiency of the key generation process depends on the method of channel measurement and quantisation.
Instead of quantising the channel measurements directly, if their reciprocity is enhanced and then quantised appropriately, the key generation process can be made efficient and fast. In this thesis, four methods of enhancing reciprocity are presented namely; l1-norm minimisation, Hierarchical clustering, Kalman filtering,
and Polynomial regression. They are appropriately quantised by binary and adaptive quantisation. Then, the entire process of key generation, from measuring the channel profile to obtaining a secure key is validated by using real-world channel measurements. The performance evaluation is done by comparing their performance in terms of bit disagreement rate, key generation rate, test of randomness,
robustness test, and eavesdropper test. An architecture, KeyBunch, for effectively
deploying the physical layer security in mobile and vehicular ad-hoc networks is
also proposed. Finally, as an use-case, KeyBunch is deployed in a secure vehicular communication architecture, to highlight the advantages offered by physical layer security.
Since their invention in the 1980s, behaviour-based systems have become very popular among roboticists. Their component-based nature facilitates the distributed implementation of systems, fosters reuse, and allows for early testing and integration. However, the distributed approach necessitates the interconnection of many components into a network in order to realise complex functionalities. This network is crucial to the correct operation of the robotic system. There are few sound design techniques for behaviour networks, especially if the systems shall realise task sequences. Therefore, the quality of the resulting behaviour-based systems is often highly dependant on the experience of their developers.
This dissertation presents a novel integrated concept for the design and verification of behaviour-based systems that realise task sequences. Part of this concept is a technique for encoding task sequences in behaviour networks. Furthermore, the concept provides guidance to developers of such networks. Based on a thorough analysis of methods for defining sequences, Moore machines have been selected for representing complex tasks. With the help of the structured workflow proposed in this work and the developed accompanying tool support, Moore machines defining task sequences can be transferred automatically into corresponding behaviour networks, resulting in less work for the developer and a lower risk of failure.
Due to the common integration of automatically and manually created behaviour-based components, a formal analysis of the final behaviour network is reasonable. For this purpose, the dissertation at hand presents two verification techniques and justifies the selection of model checking. A novel concept for applying model checking to behaviour-based systems is proposed according to which behaviour networks are modelled as synchronised automata. Based on such automata, properties of behaviour networks that realise task sequences can be verified or falsified. Extensive graphical tool support has been developed in order to assist the developer during the verification process.
Several examples are provided in order to illustrate the soundness of the presented design and verification techniques. The applicability of the integrated overall concept to real-world tasks is demonstrated using the control system of an autonomous bucket excavator. It can be shown that the proposed design concept is suitable for developing complex sophisticated behaviour networks and that the presented verification technique allows for verifying real-world behaviour-based systems.
This thesis deals with the development of a tractor front loader scale which measures payload continuously, independent of the center of gravity of the payload, and unaffected of the position and movements of the loader. To achieve this, a mathematic model of a common front loader is simplified which makes it possible to identify its parameters by a repeatable and automatic procedure. By measuring accelerations as well as cylinder forces, the payload is determined continuously during the working process. Finally, a prototype was build and the scale was tested on a tractor.
Today’s pervasive availability of computing devices enabled with wireless communication and location- or inertial sensing capabilities is unprecedented. The number of smartphones sold worldwide are still growing and increasing numbers of sensor enabled accessories are available which a user can wear in the shoe or at the wrist for fitness tracking, or just temporarily puts on to measure vital signs. Despite this availability of computing and sensing hardware the merit of application seems rather limited regarding the full potential of information inherent to such senor deployments. Most applications build upon a vertical design which encloses a narrowly defined sensor setup and algorithms specifically tailored to suit the application’s purpose. Successful technologies, however, such as the OSI model, which serves as base for internet communication, have used a horizontal design that allows high level communication protocols to be run independently from the actual lower-level protocols and physical medium access. This thesis contributes to a more horizontal design of human activity recognition systems at two stages. First, it introduces an integrated toolchain to facilitate the entire process of building activity recognition systems and to foster sharing and reusing of individual components. At a second stage, a novel method for automatic integration of new sensors to increase a system’s performance is presented and discussed in detail.
The integrated toolchain is built around an efficient toolbox of parametrizable components for interfacing sensor hardware, synchronization and arrangement of data streams, filtering and extraction of features, classification of feature vectors, and interfacing output devices and applications. The toolbox emerged as open-source project through several research projects and is actively used by research groups. Furthermore, the toolchain supports recording, monitoring, annotation, and sharing of large multi-modal data sets for activity recognition through a set of integrated software tools and a web-enabled database.
The method for automatically integrating a new sensor into an existing system is, at its core, a variation of well-established principles of semi-supervised learning: (1) unsupervised clustering to discover structure in data, (2) assumption that cluster membership is correlated with class membership, and (3) obtaining at a small number of labeled data points for each cluster, from which the cluster labels are inferred. In most semi-supervised approaches, however, the labels are the ground truth provided by the user. By contrast, the approach presented in this thesis uses a classifier trained on an N-dimensional feature space (old classifier) to provide labels for a few points in an (N+1)-dimensional feature space which are used to generate a new, (N+1)-dimensional classifier. The different factors that make a distribution difficult to handle are discussed, a detailed description of heuristics designed to mitigate the influences of such factors is provided, and a detailed evaluation on a set of over 3000 sensor combinations from 3 multi-user experiments that have been used by a variety of previous studies of different activity recognition methods is presented.
Large displays become more and more popular, due to dropping prices. Their size and high resolution leverages collaboration and they are capable of dis- playing even large datasets in one view. This becomes even more interesting as the number of big data applications increases. The increased screen size and other properties of large displays pose new challenges to the Human- Computer-Interaction with these screens. This includes issues such as limited scalability to the number of users, diversity of input devices in general, leading to increased learning efforts for users, and more.
Using smart phones and tablets as interaction devices for large displays can solve many of these issues. Since they are almost ubiquitous today, users can bring their own device. This approach scales well with the number of users. These mobile devices are easy and intuitive to use and allow for new interaction metaphors, as they feature a wide array of input and output capabilities, such as touch screens, cameras, accelerometers, microphones, speakers, Near-Field Communication, WiFi, etc.
This thesis will present a concept to solve the issues posed by large displays. We will show proofs-of-concept, with specialized approaches showing the via- bility of the concept. A generalized, eyes-free technique using smart phones or tablets to interact with any kind of large display, regardless of hardware or software then overcomes the limitations of the specialized approaches. This is implemented in a large display application that is designed to run under a multitude of environments, including both 2D and 3D display setups. A special visualization method is used to combine 2D and 3D data in a single visualization.
Additionally the thesis will present several approaches to solve common is- sues with large display interaction, such as target sizes on large display getting too small, expensive tracking hardware, and eyes-free interaction through vir- tual buttons. These methods provide alternatives and context for the main contribution.
In this thesis we extend the worst-case modeling approach as first introduced by Hua and Wilmott (1997) (option pricing in discrete time) and Korn and Wilmott (2002) (portfolio optimization in continuous time) in various directions.
In the continuous-time worst-case portfolio optimization model (as first introduced by Korn and Wilmott (2002)), the financial market is assumed to be under the threat of a crash in the sense that the stock price may crash by an unknown fraction at an unknown time. It is assumed that only an upper bound on the size of the crash is known and that the investor prepares for the worst-possible crash scenario. That is, the investor aims to find the strategy maximizing her objective function in the worst-case crash scenario.
In the first part of this thesis, we consider the model of Korn and Wilmott (2002) in the presence of proportional transaction costs. First, we treat the problem without crashes and show that the value function is the unique viscosity solution of a dynamic programming equation (DPE) and then construct the optimal strategies. We then consider the problem in the presence of crash threats, derive the corresponding DPE and characterize the value function as the unique viscosity solution of this DPE.
In the last part, we consider the worst-case problem with a random number of crashes by proposing a regime switching model in which each state corresponds to a different crash regime. We interpret each of the crash-threatened regimes of the market as states in which a financial bubble has formed which may lead to a crash. In this model, we prove that the value function is a classical solution of a system of DPEs and derive the optimal strategies.
Specification of asynchronous circuit behaviour becomes more complex as the
complexity of today’s System-On-a-Chip (SOC) design increases. This also causes
the Signal Transition Graphs (STGs) – interpreted Petri nets for the specification
of asynchronous circuit behaviour – to become bigger and more complex, which
makes it more difficult, sometimes even impossible, to synthesize an asynchronous
circuit from an STG with a tool like petrify [CKK+96] or CASCADE [BEW00].
It has, therefore, been suggested to decompose the STG as a first step; this
leads to a modular implementation [KWVB03] [KVWB05], which can reduce syn-
thesis effort by possibly avoiding state explosion or by allowing the use of library
elements. A decomposition approach for STGs was presented in [VW02] [KKT93]
[Chu87a]. The decomposition algorithm by Vogler and Wollowski [VW02] is based
on that of Chu [Chu87a] but is much more generally applicable than the one in
[KKT93] [Chu87a], and its correctness has been proved formally in [VW02].
This dissertation begins with Petri net background described in chapter 2.
It starts with a class of Petri nets called a place/transition (P/T) nets. Then
STGs, the subclass of P/T nets, is viewed. Background in net decomposition
is presented in chapter 3. It begins with the structural decomposition of P/T
nets for analysis purposes – liveness and boundedness of the net. Then STG
decomposition for synthesis from [VW02] is described.
The decomposition method from [VW02] still could be improved to deal with
STGs from real applications and to give better decomposition results. Some
improvements for [VW02] to improve decomposition result and increase algorithm
efficiency are discussed in chapter 4. These improvement ideas are suggested in
[KVWB04] and some of them are have been proved formally in [VK04].
The decomposition method from [VW02] is based on net reduction to find
an output block component. A large amount of work has to be done to reduce
an initial specification until the final component is found. This reduction is not
always possible, which causes input initially classified as irrelevant to become
relevant input for the component. But under certain conditions (e.g. if structural
auto-conflicts turn out to be non-dynamic) some of them could be reclassified as
irrelevant. If this is not done, the specifications become unnecessarily large, which
intern leads to unnecessarily large implemented circuits. Instead of reduction, a
new approach, presented in chapter 5, decomposes the original net into structural
components first. An initial output block component is found by composing the
structural components. Then, a final output block component is obtained by net
reduction.
As we cope with the structure of a net most of the time, it would be useful
to have a structural abstraction of the net. A structural abstraction algorithm
[Kan03] is presented in chapter 6. It can improve the performance in finding an
output block component in most of the cases [War05] [Taw04]. Also, the structure
net is in most cases smaller than the net itself. This increases the efficiency of the
decomposition algorithm because it allows the transitions contained in a node of
the structure graph to be contracted at the same time if the structure graph is
used as internal representation of the net.
Chapter 7 discusses the application of STG decomposition in asynchronous
circuit design. Application to speed independent circuits is discussed first. Af-
ter that 3D circuits synthesized from extended burst mode (XBM) specifications
are discussed. An algorithm for translating STG specifications to XBM specifi-
cations was first suggested by [BEW99]. This algorithm first derives the state
machine from the STG specification, then translates the state machine to XBM
specification. An XBM specification, though it is a state machine, allows some
concurrency. These concurrencies can be translated directly, without deriving
all of the possible states. An algorithm which directly translates STG to XBM
specifications, is presented in chapter 7.3.1. Finally DESI, a tool to decompose
STGs and its decomposition results are presented.
Industrial design has a long history. With the introduction of Computer-Aided Engineering, industrial design was revolutionised. Due to the newly found support, the design workflow changed, and with the introduction of virtual prototyping, new challenges arose. These new engineering problems have triggered
new basic research questions in computer science.
In this dissertation, I present a range of methods which support different components of the virtual design cycle, from modifications of a virtual prototype and optimisation of said prototype, to analysis of simulation results.
Starting with a virtual prototype, I support engineers by supplying intuitive discrete normal vectors which can be used to interactively deform the control mesh of a surface. I provide and compare a variety of different normal definitions which have different strengths and weaknesses. The best choice depends on
the specific model and on an engineer’s priorities. Some methods have higher accuracy, whereas other methods are faster.
I further provide an automatic means of surface optimisation in the form of minimising total curvature. This minimisation reduces surface bending, and therefore, it reduces material expenses. The best results can be obtained for analytic surfaces, however, the technique can also be applied to real-world examples.
Moreover, I provide engineers with a curvature-aware technique to optimise mesh quality. This helps to avoid degenerated triangles which can cause numerical issues. It can be applied to any component of the virtual design cycle: as a direct modification of the virtual prototype (depending on the surface defini-
tion), during optimisation, or dynamically during simulation.
Finally, I have developed two different particle relaxation techniques that both support two components of the virtual design cycle. The first component for which they can be used is discretisation. To run computer simulations on a model, it has to be discretised. Particle relaxation uses an initial sampling,
and it improves it with the goal of uniform distances or curvature-awareness. The second component for which they can be used is the analysis of simulation results. Flow visualisation is a powerful tool in supporting the analysis of flow fields through the insertion of particles into the flow, and through tracing their movements. The particle seeding is usually uniform, e.g. for an integral surface, one could seed on a square. Integral surfaces undergo strong deformations, and they can have highly varying curvature. Particle relaxation redistributes the seeds on the surface depending on surface properties like local deformation or curvature.