A new method for efficiently converting electron backscatter diffraction data obtained using serial sectioning by focused ion beam of a polycrystalline thin film into a computational, three-dimensional (3D) structure is presented. The reported data processing method results in a more accurate representation of the grain surfaces, reduced computer memory usage, and improved processing speed compared to traditional voxel methods. The grain structure of a polycrystalline absorption layer from a high-efficiency Cu(In,Ga)Se2 solar cell (19.5%) is reconstructed in 3D and the grain size and surface distribution is investigated. The grain size distribution is found to be best fitted by a log-normal distribution. We further find that the grain size is determined by the [Ga]/([Ga] + [In]) ratio in vertical direction, which was measured by glow discharge optical emission spectroscopy. Finally, the 3D model derived from the structural information is applied in optoelectronic simulations, revealing insights into the effects of grain boundary recombination on the open-circuit voltage of the solar cell. An accurate 3D structure like the one obtained with our method is a prerequisite for a detailed understanding of mechanical properties and for advanced optical and electronic simulations of polycrystalline thin films.

Leaf epidermis pavement cells form highly complex shapes with interlocking lobes and necks at their anticlinal walls. The microtubule cytoskeleton plays essential roles in pavement cell morphogenesis, in particular at necks. Vice versa, shape generates stress patterns that regulate microtubule organization. Genetic or pharmacological perturbations that affect pavement cell shape often affect microtubule organization. Pavement cell shape and microtubule organization are therefore closely interconnected. Here, we present commonly used approaches for the quantitative analysis of pavement cell shape characteristics and of microtubule organization. In combination with ablation experiments, these methods can be applied to investigate how different genotypes (or treatments) affect the organization and stress responsiveness of the microtubule cytoskeleton.

This habilitation thesis presents methodical and computational approaches for the analysis of image data in various application domains of the life sciences. Amongst others methods for the detection of punctiform structures in microscope images are described. A second research field deals with the segmentation of more complex objects like cells or plant roots. Here established methods like active contours have specifically been extended as well as new methods of machine learning with neural networks been applied. A third topic area comprises the extraction of quantitative data of objects. Here new measures for cell shape have been devised as well as protocols for the quantification and comparison of subcellular structures been developed. In addition to the methodical approaches the software libraries Alida and MiToBo are presented, which yield the basis for the implementation of most of the approaches, and the annotation tool rhizoTrak.

Roots and their temporal development play an important role in plant research. Over the decades image-based monitoring of root growth has become a key methodology in this research field. The growing amount of image data is often tackled with automatic image analysis approaches. In particular convolutional neural networks (CNNs) recently gained increasing interest for root segmentation. This segmentation of roots is usually only the first step of an analysis pipeline and needs to be supplemented by topological reconstruction of the complete root system architecture.
In this paper we present a comprehensive study of different CNN architectures, loss functions and parameter settings for root image segmentation. In addition, we show how main and lateral roots can be identified based on the skeletons of segmented root components as a first step towards topological reconstruction of root system architecture. We present quantitative and qualitative results on data released in the course of the CVPPA Arabidopsis Root Segmentation Challenge 2021.

Leaf epidermis pavement cells develop complex jigsaw puzzle-like shapes in many plant species, including the model plant Arabidopsis thaliana. Due to their complex morphology, pavement cells have become a popular model system to study shape formation and coordination of growth in the context of mechanically coupled cells at the tissue level. To facilitate robust assessment and analysis of pavement cell shape characteristics in a high-throughput fashion, we have developed PaCeQuant and a collection of supplemental tools. The ImageJ-based MiToBo plugin PaCeQuant supports fully automatic segmentation of cell contours from microscopy images and the extraction of 28 shape features for each detected cell. These features now also include the Largest Empty Circle criterion as a proxy for mechanical stress. In addition, PaCeQuant provides a set of eight features for individual lobes, including the categorization as type I and type II lobes at two- and three-cell junctions, respectively. The segmentation and feature extraction results of PaCeQuant depend on the quality of input images. To allow for corrections in case of local segmentation errors, the LabelImageEditor is provided for user-friendly manual postprocessing of segmentation results. For statistical analysis and visualization, PaCeQuant is supplemented with the R package PaCeQuantAna, which provides statistical analysis functions and supports the generation of publication-ready plots in ready-to-use R workflows. In addition, we recently released the FeatureColorMapper tool which overlays feature values over cell regions for user-friendly visual exploration of selected features in a set of analyzed cells.

Background and aims. Minirhizotrons are commonly used to study root turnover which is essential for understanding ecosystem carbon and nutrient cycling. Yet, extracting data from minirhizotron images requires extensive annotation effort. Existing annotation tools often lack flexibility and provide only a subset of the required functionality. To facilitate efficient root annotation in minirhizotrons, we present the user-friendly open source tool rhizoTrak.

Methods and results. rhizoTrak builds on TrakEM2 and is publicly available as Fiji plugin. It uses treelines to represent branching structures in roots and assigns customizable status labels per root segment. rhizoTrak offers configuration options for visualization and various functions for root annotation mostly accessible via keyboard shortcuts. rhizoTrak allows time-series data import and particularly supports easy handling and annotation of time-series images. This is facilitated via explicit temporal links (connectors) between roots which are automatically generated when copying annotations from one image to the next. rhizoTrak includes automatic consistency checks and guided procedures for resolving inconsistencies. It facilitates easy data exchange with other software by supporting open data formats.

Conclusions. rhizoTrak covers the full range of functions required for user-friendly and efficient annotation of time-series images. Its flexibility and open source nature will foster efficient data acquisition procedures in root studies using minirhizotrons.

The microtubule cytoskeleton plays important roles in cell morphogenesis. To investigate the mechanisms of cytoskeletal organization, for example, during growth or development, in genetic studies, or in response to environmental stimuli, image analysis tools for quantitative assessment are needed. Here, we present a method for texture measure-based quantification and comparative analysis of global microtubule cytoskeleton patterns and subsequent visualization of output data. In contrast to other approaches that focus on the extraction of individual cytoskeletal fibers and analysis of their orientation relative to the growth axis, CytoskeletonAnalyzer2D quantifies cytoskeletal organization based on the analysis of local binary patterns. CytoskeletonAnalyzer2D thus is particularly well suited to study cytoskeletal organization in cells where individual fibers are difficult to extract or which lack a clearly defined growth axis, such as leaf epidermal pavement cells. The tool is available as ImageJ plugin and can be combined with publicly available software and tools, such as R and Cytoscape, to visualize similarity networks of cytoskeletal patterns.

Morphological analysis of cell shapes requires segmentation of cell contours from input images and subsequent extraction of meaningful shape descriptors that provide the basis for qualitative and quantitative assessment of shape characteristics. Here, we describe the publicly available ImageJ plugin PaCeQuant and its associated R package PaCeQuantAna, which provides a pipeline for fully automatic segmentation, feature extractionFeature extraction, statistical analysis, and graphical visualization of cell shape properties. PaCeQuant is specifically well suited for analysis of jigsaw puzzle-like leaf epidermis pavement cells from 2D input images and supports the quantification of global, contour-based, skeleton-based, and pavement cell-specific shape descriptors.

The structure of the microtubule cytoskeleton provides valuable information related to morphogenesis of cells. The cytoskeleton organizes into diverse patterns that vary in cells of different types and tissues, but also within a single tissue. To assess differences in cytoskeleton organization methods are needed that quantify cytoskeleton patterns within a complete cell and which are suitable for large data sets. A major bottleneck in most approaches, however, is a lack of techniques for automatic extraction of cell contours. Here, we present a semi-automatic pipeline for cell segmentation and quantification of microtubule organization. Automatic methods are applied to extract major parts of the contours and a handy image editor is provided to manually add missing information efficiently. Experimental results prove that our approach yields high-quality contour data with minimal user intervention and serves a suitable basis for subsequent quantitative studies.

Plastids are involved in many fundamental biochemical pathways in plants. They can produce tubular membrane out-folds from their surface. These so-called stromules have initially been described over a century ago, but their functional role is still elusive. To identify cellular processes or genetic elements underlying stromule formation screens of large populations of mutant plants or plants under different treatments are carried out and stromule frequencies are extracted. Due to a lack of automatic methods, however, this quantification is usually done manually rendering this step a main bottleneck in stromule research. Here, we present a new approach for quantification of stromule frequencies. Plastids are extracted from microscope images using local wavelet analysis over multiple scales combined with statistical hypothesis testing to resolve competing detections from different scales. Subsequently, for each plastid region evidence for the existence of stromules in its vicinity is investigated applying ridge detection techniques and geometric criteria. Experimental results prove that our approach is suitable to properly identify stromules. Even in microscopy images with a high noise level and distracting signals extracted stromule counts are comparable to those of biological experts.

Plant microtubules form a highly dynamic intracellular network with important roles for regulating cell division, cell proliferation, and cell morphology. Their organization and dynamics are co-ordinated by various microtubule-associated proteins (MAPs) that integrate environmental and developmental stimuli to fine-tune and adjust cytoskeletal arrays. IQ67 DOMAIN (IQD) proteins recently emerged as a class of plant-specific MAPs with largely unknown functions. Here, using a reverse genetics approach, we characterize Arabidopsis IQD5 in terms of its expression domains, subcellular localization, and biological functions. We show that IQD5 is expressed mostly in vegetative tissues, where it localizes to cortical microtubule arrays. Our phenotypic analysis of iqd5 loss-of-function lines reveals functions of IQD5 in pavement cell (PC) shape morphogenesis. Histochemical analysis of cell wall composition further suggests reduced rates of cellulose deposition in anticlinal cell walls, which correlate with reduced anisotropic expansion. Lastly, we demonstrate IQD5-dependent recruitment of calmodulin calcium sensors to cortical microtubule arrays and provide first evidence for important roles for calcium in regulation of PC morphogenesis. Our work identifies IQD5 as a novel player in PC shape regulation and, for the first time, links calcium signaling to developmental processes that regulate anisotropic growth in PCs.

The morphology of cells varies between organs, tissues, and even within a single tissue, ranging from simple to highly complex shapes. Morphology contributes to cell functions and its analysis allows to identify the mechanisms that drive specialization. Robust quantification demands for analyzing large data sets and raises the need for unbiased automatic analysis algorithms that quantify shape features to reflect cellular complexity. Pavement cells are the most abundant cell type of the leaf epidermis and form one of the most complex cell types in plants. Quantifying their jigsaw puzzle-like shapes with interdigitated lobes and necks renders a major challenge. To address this challenge we present PaCeQuant, an ImageJ plugin for high throughput cell segmentation and shape quantification from fluorescence microscopy images. The plugin is part of MiToBo and combines algorithms for automatic segmentation of cell contours with extraction of a flexible set of shape features for individual cells. Cell boundary segmentation relies on vesselness enhancement filters followed by local binarization and morphological post-processing. Remaining gaps are closed by exploiting results of a binary watershed transformation. Extracted shape features cover a large collection of potentially interesting cell characteristics and subsume unique pavement cell-specific features like lobe and neck count, neck width or non-lobe area. Moreover, PaCeQuant is the first tool supporting the analysis of individual lobes. The core plugin is supplemented with an R script for detailed statistical evaluation, visualization and comparative analysis of extracted data, particularly helpful in identifying suitable features for specific research questions. In future PaCeQuant's segmentation algorithms will be extended towards additional types of images and larger robustness against low image quality, and an R package will be released to further ease interpretation of shape features.

Minirhizotrons are widely used to study the dynamics of plant root systems in a non-destructive way. The time series images acquired contain valuable information about root processes over time. However, annotation of roots usually requires time-consuming manual work due to the complexity of the images. To ease this annotation process we present the open source software rhizoTrak. rhizoTrak builds on the Fiji plugin TrakEM2 for neural circuit reconstruction. All images of a time series can be imported at once and easily be browsed and compared. For annotation of roots, rhizoTrak adopts the treeline data type of TrakEM2 allowing to represent complete root systems. In addition, rhizoTrak inherits manifold edit operations for treelines from TrakEM2, like moving, adding or deleting nodes, assigning labels to individual segments, or joining and splitting treelines. rhizoTrak provides an extended set of keyboard shortcuts and allows for a configurable set of strings as segment labels. For estimating root dynamics it is essential to identify the same physical root in all images of a series. rhizoTrak adopts the connector concept of TrakEM2 to link treeline annotations in different images. However, in rhizoTrak connector membership is based on object identity rather than geometric position which avoids breaking membership if the geometry of a treeline is modified. Root annotations can be propagated from one image to the next with connectors being automatically established. Merge operations on treelines can cause topological inconsistencies if involved treelines are members of different connectors. To support the user in handling such situations a conflict editor is available. rhizoTrak consequently relies on open data formats, like RSML for exchange of annotation data, and allows to export detailed root statistics. In near future automatic segmentation algorithms will be integrated to further reduce the effort of annotating minirhizotron images.

Pavement cells (PCs) are the most frequently occurring cell type in the leaf epidermis and play important roles in leaf growth and function. In many plant species, PCs form highly complex jigsaw-puzzle-shaped cells with interlocking lobes. Understanding of their development is of high interest for plant science research because of their importance for leaf growth and hence for plant fitness and crop yield. Studies of PC development, however, are limited, because robust methods are lacking that enable automatic segmentation and quantification of PC shape parameters suitable to reflect their cellular complexity. Here, we present our new ImageJ-based tool, PaCeQuant, which provides a fully automatic image analysis workflow for PC shape quantification. PaCeQuant automatically detects cell boundaries of PCs from confocal input images and enables manual correction of automatic segmentation results or direct import of manually segmented cells. PaCeQuant simultaneously extracts 27 shape features that include global, contour-based, skeleton-based, and PC-specific object descriptors. In addition, we included a method for classification and analysis of lobes at two-cell junctions and three-cell junctions, respectively. We provide an R script for graphical visualization and statistical analysis. We validated PaCeQuant by extensive comparative analysis to manual segmentation and existing quantification tools and demonstrated its usability to analyze PC shape characteristics during development and between different genotypes. PaCeQuant thus provides a platform for robust, efficient, and reproducible quantitative analysis of PC shape characteristics that can easily be applied to study PC development in large data sets.

Data analysis procedures can often be modeled as a set of manipulation operations applied to input data and resulting in transformed intermediate and result data. The Java library Alida is providing an advanced development framework to support programmers in developing data analysis applications adhering to such a scheme. The main intention of Alida is to foster re-usability by offering well-defined, unified, modular APIs and execution procedures for operators, and to ease development by releasing developers from tedious tasks. Alida features automatic generation of handy graphical and command line user interfaces, a built-in graphical editor for workflow design, and an automatic documentation of analysis pipelines. Alida is available from its project webpage http://www.informatik.uni-halle.de/alida, on Github and via our Maven server.

Calcium (Ca2+) signaling and dynamic reorganization of the cytoskeleton are essential processes for the coordination and control of plant cell shape and cell growth. Calmodulin (CaM) and closely related calmodulin-like (CML) polypeptides are principal sensors of Ca2+ signals. CaM/CMLs decode and relay information encrypted by the second messenger via differential interactions with a wide spectrum of targets to modulate their diverse biochemical activities. The plant-specific IQ67 DOMAIN (IQD) family emerged as possibly the largest class of CaM-interacting proteins with undefined molecular functions and biological roles. Here, we show that the 33 members of the IQD family in Arabidopsis (Arabidopsis thaliana) differentially localize, using green fluorescent protein (GFP)-tagged proteins, to multiple and distinct subcellular sites, including microtubule (MT) arrays, plasma membrane subdomains, and nuclear compartments. Intriguingly, the various IQD-specific localization patterns coincide with the subcellular patterns of IQD-dependent recruitment of CaM, suggesting that the diverse IQD members sequester Ca2+-CaM signaling modules to specific subcellular sites for precise regulation of Ca2+-dependent processes. Because MT localization is a hallmark of most IQD family members, we quantitatively analyzed GFP-labeled MT arrays in Nicotiana benthamiana cells transiently expressing GFP-IQD fusions and observed IQD-specific MT patterns, which point to a role of IQDs in MT organization and dynamics. Indeed, stable overexpression of select IQD proteins in Arabidopsis altered cellular MT orientation, cell shape, and organ morphology. Because IQDs share biochemical properties with scaffold proteins, we propose that IQD families provide an assortment of platform proteins for integrating CaM-dependent Ca2+ signaling at multiple cellular sites to regulate cell function, shape, and growth.

MiToBo is a toolbox and Java library for solving basic as well as advanced image processing and analysis tasks. It features a rich collection of fundamental, intermediate and high-level image processing operators and algorithms as well as a couple of sophisticated tools for specific biological and biomedical applications. These tools include operators for elucidating cellular morphology and locomotion as well as operators for the characterization of certain intracellular particles and structures.

MiToBo builds upon and integrates into the widely-used image analysis software packages ImageJ and Fiji [11, 10], and all of its operators can easily be run in ImageJ and Fiji via a generic operator runner plugin. Alternatively MiToBo operators can directly be run from command line, and using its functionality as a library for developing own applications is also supported. Thanks to the Alida library [8] forming the base of MiToBo all operators share unified APIs fostering reusability, and graphical as well as command line user interfaces for operators are automatically generated. MiToBo is available from its website http://www.informatik.uni-halle.de/mitobo, on Github, via an Apache Archiva Maven repository server, and it can easily be activated in Fiji via its own update site.

Solving image analysis problems is not restricted to the pure delineation of algorithms suitable to tackle the task at hand. Rather these also need to be made available to the users promptly and equipped with handy user interfaces to foster progress in the intended field of application. Alida is a software framework to advance the integrated development of algorithms and appropriate user interfaces. It automatically generates user interfaces for implemented algorithms, offers an automatic documentation of analysis procedures, and ships with a graphical editor for designing complex workflows. Alida's Java implementation is licensed under GPL 3.0 and publicly available at http://www.informatik.uni-halle.de/alida.

Organelle morphology as well as subcellular organisation can drastically change in response to a changing cellular environment. However our knowledge about the functionality as well as the regulation of organelle form changes, such as plastid tubule formation (stromules) and organelle rearrangements, is limited. Monitoring changes to organelle morphology and subcellular organisation in response to experimental treatments and in different mutant backgrounds is a promising strategy to address open questions. However, for a detailed comparison of different treatments and mutants, the quantification of subcellular phenotypes is crucial. Unfortunately, for many specific subcellular parameters, such as stromule frequency, software tools supporting data extraction from images are not readily available or are highly specialized in their purpose. In order to quantify stromule frequency in a semi-automated manner we developed the 'MTB Cell Counter', which combines automated plastid detection with manual counting tools. We show that with the use of our plugin stromule frequency can be quantified up to 90% faster. We further demonstrate, with the detection of peroxisomes and nuclei, that due to its adaptable detection algorithm, which is based on scale-adaptive analysis of wavelet coefficients, the plugin can be used to reliably detect and count organelles of different size and brightness. In addition to the analysis of CLSM images, the 'MTB Cell Counter' is easily adapted to particle detection in challenging epifluorescence images, making it a versatile, semi-automated tool capable of quantifying a wide variety of subcellular phenotypes.

During cellular stress, protein synthesis is severely reduced and bulk mRNA is recruited to stress granules (SGs). Previously, we showed that the SG-recruited IGF2 mRNA-binding protein 1 (IGF2BP1) interferes with target mRNA degradation during cellular stress. Whether this requires the formation of SGs remained elusive. Here, we demonstrate that the sustained inhibition of visible SGs requires the concomitant knockdown of TIA1, TIAR and G3BP1. FRAP and photo-conversion studies, however, indicate that these proteins only transiently associate with SGs. This suggests that instead of forming a rigid scaffold for mRNP recruitment, TIA proteins and G3BP1 promote SG-formation by constantly replenishing mRNPs. In contrast, RNA-binding proteins like IGF2BP1 or HUR, which are dispensable for SG-assembly, are stably associated with SGs and the IGF2BP1/HUR-G3BP1 association is increased during stress. The depletion of IGF2BP1 enhances the degradation of target mRNAs irrespective of inhibiting SG-formation, whereas the turnover of bulk mRNA remains unaffected when SG-formation is impaired. Together these findings indicate that the stabilization of mRNAs during cellular stress is facilitated by the formation of stable mRNPs, which are recruited to SGs by TIA proteins and/or G3BP1. Importantly, however, the aggregation of mRNPs to visible SGs is dispensable for preventing mRNA degradation.

The analysis of F-actin organization in cells is a key topic in many fields of biomedical research. While standard protocols for imaging immunostained actin are well established, assessment of the resulting microscopy images is most of the time still performed manually and with a high degree of subjectivity. In this paper, we present a new approach for the analysis of actin structures in microscopy images and the quantification of differences and similarities in actin organization between cells. Compared to existing methods, our approach does not require any previous knowledge about the cells or structures to be analyzed. It works in an unsupervised fashion, combining statistical texture measures and clustering techniques. By this, our method yields large flexibility and allows for application in a wide range of ex- perimental scenarios, and also to heterogeneous cell populations. Experimental evaluation on sample data proves that our method yields meaningful results for biomedical investigations.

The function of complex networks in the nervous system relies on the proper formation of neuronal contacts and their remodeling. To decipher the molecular mechanisms underlying these processes, it is essential to establish unbiased automated tools allowing the correlation of neurite morphology and the subcellular distribution of molecules by quantitative means. We developed NeuronAnalyzer2D, a plugin for ImageJ, which allows the extraction of neuronal cell morphologies from two dimensional high resolution images, and in particular their correlation with protein profiles determined by indirect immunostaining of primary neurons. The prominent feature of our approach is the ability to extract subcellular distributions of distinct biomolecules along neurites. To extract the complete areas of neurons, required for this analysis, we employ active contours with a new distance based energy. For locating the structural parts of neurons and various morphological parameters we adopt a wavelet based approach. The presented approach is able to extract distinctive profiles of several proteins and reports detailed morphology measurements on neurites. We compare the detected neurons from NeuronAnalyzer2D with those obtained by NeuriteTracer and Vaa3D-Neuron, two popular tools for automatic neurite tracing. The distinctive profiles extracted for several proteins, for example, of the mRNA binding protein ZBP1, and a comparative evaluation of the neuron segmentation results proves the high quality of the quantitative data and proves its practical utility for biomedical analyses.

Solving image analysis problems is not restricted to the pure delineation of algorithms suitable to tackle the task at hand. Rather these also need to be made available to the users promptly and equipped with handy user interfaces to foster progress in the intended field of application. Alida is a software framework to advance the integrated development of algorithms and appropriate user interfaces. It automatically generates user interfaces for implemented algorithms, offers an automatic documentation of analysis procedures, and ships with a graphical editor for designing complex workflows. Alida’s Java implementation is licensed under GPL 3.0 and publicly available at http://www.informatik.uni-halle.de/alida.

For multi-view 3D reconstruction robust standard procedures have been established and can directly be applied to many scenarios. However, the extraction of point correspondences as a prerequisite for reconstruction is demanding for various applications. Here we present a new analysis pipeline for 3D reconstruction in the field of barley plant monitoring. Barley plants show a significant structural and textural similarity rendering the application of standard procedures to extract correspondences impossible. Our new approach overcomes these problems by combining information from various cues over different stages. Experiments on real data prove the suitability of our approach to generate 3D models of the plants from which phenotypical data can easily be derived.

Solving image analysis problems not only requires the development of suitable sets of algorithms to produce desired result data, but also demands for suitable user interfaces (UIs) to foster use in practice. Here we present our library Alida which aims to promote the deployment of UIs by featuring their automatic generation from algorithm code. Alida supports command line and graphical UIs (GUIs), and ships with a graphical editor for designing more complex workflows. Enforcing only a small set of rules to obey Alida significantly reduces implementation overhead for developers and allows for focusing on algorithm rather than UI design. The suitability of Alida’s concept for real-life applications is shown by the library MiToBo for biomedical image analysis implemented based on Alida.

Scratch assays are a widely used technique for assessing the migratory potential of cells. However, due to a lack of appropriate tools these assays are often analyzed manually. Here we present a plugin for the automatic segmentation and analysis of scratch assay images. The input images are segmented using a topology-preserving non-PDE level set approach. To exclude images where the scratch has already been completely closed a support vector machine separates images containing a scratch from such that do not. As our method requires adjustment of only two parameters, it is easy to use and has already been successfully applied in practice.

Solving challenging image analysis problems usually requires to combine several individual analysis steps into comprehensive workflows. To find a suitable combination of algorithms is often quite elaborate and accomplished interactively. In this paper we present our graphical program editor ”Grappa” (Graphical Program Editor for Alida) supporting the design of workflows from individual image analysis operators in a graphical manner. Grappa is built on top of Alida and takes full advantage of Alida's operator annotation and generic execution mechanisms, as well as of its functionality to automatically generate graphical user interfaces for operators. Grappa is under active development and available as prototypical plugin for ImageJ and ImageJ 2.0.

Parametric active contour models, commonly denoted as snakes, provide an integrated framework for transform- ing object segmentation tasks into well-founded mathematical optimization problems. Over the years snakes have proven large flexibility in segmenting a great variety of different objects in lots of setups. Nevertheless, each new object segmentation problem again raises the question if snakes are capable of solving the problem and if so, which combination of energies to apply. Here we present our ImageJ plugin 'SnakeOptimizer' which implements object segmentation based on 2D parametric active contours. It is well-suited to support users in answering the above questions during rapid-prototyping stages, as well as for being integrated into productive analysis workflows. The functional core of the implementation can easily be extended with new energy models based on functionality provided by Alida.1 The ImageJ plugin features a handy GUI for interactive use, and the optimizer can also be run headless from command line or on the programming level using its intuitive API.

Analysis of data is subject of many fields of applications. Besides the results per se, the documentation of analysis processes is important for later verification, reconstruction, or publication. As manual documentation is tedious and error prone the Java library Alida provides fully automatic extraction of processing histories. As applications we present our own image processing toolbox MiToBo based on ImageJ and demonstrate the integration of Alida into ImageJ 2.0.

Cell migration assessment is often done by scratch assay experiments for which quantitative evaluations are usually performed manually. Here we present an automatic analysis pipeline detecting scratch boundaries and measuring areas based on level sets. We extend non-PDE level sets for topology-preservation and use an entropy-based energy functional. This approach by design segments a scratch in every image, hence, we employ support vector machines to identify images showing no scratch at all. Compared to other algorithms our approach, implemented as ImageJ plugin, relies on a minimal set of parameters. Experimental evaluations show the high quality of results and their suitability for biomedical investigations.

Application of active contours for image segmentation raises the question of contour representation, i.e. whether to use snakes or level sets. The representation directly affects is- sues like topology-preservation and energy optimization. In this paper we aim to contribute to the understanding of spe- cific characteristics of contour representations with a detailed comparison of snakes vs. non-PDE level sets. Based on the same energy functional and applied to different kinds of real- world data our experiments show minor differences in seg- mentation quality, but outline important distinctions regarding implementation, parameter settings and computational effort.

Analysis of biomedical data may be interpreted as a flow of objects through an analysis pipeline. The Java framework Alida (http://www.informatik.uni-halle.de/alida) defines the concept of operators as the single places of these manipulations. Typically, operators may be invoked sequentially or in parallel, and often also nested. Besides invocation on the programming level their functionality should also be available directly to users, including developers of algorithms and non-experts. This calls for graphical as well as command line interfaces. Eliminating the need to explicitly code these interfaces, Alida features fully automated generation of graphical and command line user interfaces for each operator implemented in the Alida framework. The basis is a formalism for an operator to define all input and output data objects and parameters to control processing. Automatic generation of interfaces is based on the model view presenter design pattern to achieve maximal independence between the operators, interfaces, and I/O of data objects. For implementation Java's annotation mechanism is used. The programmer is only required to properly annotate classes and member variables. Out of the box this facilitates I/O for a wide variety of Java objects including primitive data types, enumerations, arrays, and collections. In a generic way Alida handles also operators as parameters of other operators and inheritance. Only specialized classes like images require additional data providers to be implemented. While Alida is devised for data processing in general, it is used in our image analysis toolbox MiToBo (http://www.informatik.uni-halle.de/mitobo) for biomedical image analysis which is based on ImageJ and compatible to it.

Quantitative analysis of microscopy cell images requires accurate detection of cell boundaries, nuclei and sub-cellular structures. Accordingly, tools for integrated cell image analysis are required which not only provide a variety of different segmentation and detection algorithms, but at the same time support easy usage also by life scientists. In this paper we present our integrated cell analysis tool MiCA offering different segmentation techniques, e.g., based on wavelets or snakes. Iterative optimization of snake energies depends on a variety of parameters that require thorough adjustment for optimal results. To facilitate easy use of these techniques MiCA provides a new energy normalization scheme for snakes allowing for intuitive interpretation of energy parameters and, thus, simplified cell image analysis. The high quality of the result data of MiCA is proven on two sample data sets by qualitative assessments and ground-truth comparisons.

Scratch assays are widely used for cell motility and migration assessment in biomedical research. However, quantitative data is very often extracted manually. Here, we present a fully automated analysis pipeline for detecting scratch boundaries and measuring areas in scratch assay images based on level set techniques. In particular, non-PDE level sets are extended for topology preservation and applied to entropy data of scratch assay microscope images. Compared to other algorithms our approach, implemented in Java as ImageJ plugin based on the extension package MiToBo, relies on a minimal set of configuration parameters. Experimental evaluations show the high-quality of extracted assay data and their suitability for biomedical investigations.

Proper archiving or later reconstruction and verification of results in data analysis requires thorough logging of all manipulative actions on the data and corresponding parameter settings. Unfortunately such documentation tasks often enforce extensive and error prone manual activities by the user. To overcome these problems we present Alida, an approach for fully automatic documentation of data analysis procedures. Based on an unified operator interface all operations on data including their sequence and configurations are registered during analysis. Subsequently these data are made explicit in XML graph representations yielding a suitable base for visual and analytic inspection. As example for the application of Alida in practice we present MiToBo, a toolbox for image analysis implemented on the basis of Alida and demonstrating the advantages of automatic documentation for image analysis procedures.

Analysis of biomolecules in cells essentially relies on fluorescence microscopy. In combination with fully automatic image analysis it allows for in- sights into biological processes on the sub-cellular level and thus provides valu- able information for systems biology studies. In this paper we present two new techniques for automatic segmentation of cell areas and included sub-cellular par- ticles. A new cascaded and intensity-adaptive segmentation scheme based on cou- pled active contours is used to segment cell areas. Structures on the sub-cellular level, i.e. stress granules and processing bodies, are detected applying a scale- adaptive wavelet-based detection technique. Combining these results yields fully automated analyses of biological processes, and allows for new insights into in- teractions between different cellular structures and their distributions among dif- ferent cells. We present an experimental evaluation based on ground-truth data that confirms the high-quality of our segmentation results regarding these aims and opens perspectives towards deeper insights into biological systems for other problems from systems biology.

Cell tissue in microscope images is often grained and its intensities do not well agree with Gaussian distribution assumptions widely used in many segmentation approaches. We present a new cascaded segmentation scheme for inhomogeneous cell tissue based on active contour models. Cell regions are iteratively expanded from initial nuclei regions applying a data-dependent number of optimization levels. Experimental results on a set of microscope images from a human hepatoma cell line prove high quality of the results with regard to the cell segmentation task and biomedical investigations.

Microscope imaging is an indispensable tool in modern systems biology. In combination with fully automatic image analysis it allows for valuable insights into biological processes on the sub-cellular level and fosters understanding of biological systems. In this paper we present two new techniques for automatic segmentation of cell areas and included sub-cellular particles. A new cascaded and intensity-adaptive segmentation scheme based on coupled active contours is used to segment cell areas. Structures on the sub-cellular level, i.e. stress granules and processing bodies, are detected applying a scale-adaptive wavelet-based detection technique. Combining these results allows for complementary analysis of biological processes. It yields new insights into interactions between different particles and distributions of particles among different cells. Our experimental evaluations based on ground-truth data prove the high-quality of our segmentation results regarding these aims and open perspectives towards deeper insights into biological systems on the sub-cellular level.

Stress granules and processing bodies play a major role in analysing the physiology of cells under various environmental conditions. We present a fully automatic approach to detect such particles in fluorescence labeled microscope images. The detection is based on scale-adaptive analysis of wavelet coefficients allowing for an accurate detection of particles with a large variety in size. Results on real images illustrate the appropriateness of our approach and proof high quality.

Microscope imaging technologies have turned out to yield an indispensable tool in modern biomedical research. Combined with fluorescence labeling techniques they not only provide new perspectives on tissues and cells as a whole, but also on processes at the cellular level, and will be one important experimental technique of systems biology. To handle this steadily increasing amount of image data, in this paper we propose a new and fully automatic approach for neurite segmentation and protein quantification. Our technique combines three phases of automatic neuron cell localization, neurite segmentation and protein analysis. In the second stage active contour models based on hierarchical Gradient Vector Flow fields are employed, enabling precise neurite segmentation despite inhomogeneous texture. Neurite segmentation results as well as protein quantification profiles from a set of test images demonstrate the appropriateness of our approach for practical biomedical research.

Proteomics is a rapidly growing field of modern biology. Since quantitative data of proteins involved in dynamic processes of living organisms are essential for understanding the basics of life, techniques like 2-DE and related procedures for automatic data interpretation are at the heart of this research field. They are strongly required to enable analysis and interpretation of the emerging amount of available data. Analyzing and interpreting gel image data usually requires the comparison of gels from different experiments and, thus, a prior registration of gels. This can be accomplished using featureless, feature-based or hybrid registration approaches combining both techniques. Recently, the latter ones have shown high performance, and it is undoubtful that in general robust and reliable features are an essential ingredient and valuable source of information for high-quality image registration. In this paper we provide a thorough overview and elaborate analysis of the capabilities of available feature detectors for gel image registration. Particularly, a detailed and extensive comparative study is presented where common spot-specific detectors are included as well as image-content independent detectors that were not applied to the task of gel image registration until now. The study incorporates tests on several thousand synthetically deformed images from different experimental conditions. As a result it provides valuable quantitative data allowing for direct objective comparisons of various detectors, and is well suited to guide the design of new registration algorithms.

Ensembles of multiple (active) cameras yield an important ingredient in modern tracking and surveillance applications. They overcome the limited fields-of-view of single cameras, however, require robust procedures for handing over tracking tasks from one camera to another. In this paper a calibration-free procedure is proposed that allows for fast and reliable camera hand-over in Ambient Intelligence (AmI) applications. The approach is based on online acquisition of scenario-specific target models and especially solves the problem of significant changes in object view during hand-over. Real-world results acquired in an AmI environment prove the effectiveness of our technique.

When unmanned underwater vehicles (UUVs) perform missions near the ocean floor, optical sensors can be used to improve local navigation. Video mosaics allow to efficiently process the images acquired by the vehicle, and also to obtain position estimates. We discuss in this paper the role of lens distortions in this context, proving that degenerate mosaics have their origin not only in the selected motion model or in registration errors, but also in the cumulative effect of radial distortion residuals. Additionally, we present results on the accuracy of different feature-based approaches for self-correction of lens distortions that may guide the choice of appropriate techniques for correcting distortions.

In this study we consider the performance of different feature detectors used as the basis for the registration of images from two-dimensional gel electrophoresis. These are three spot detectors also used to identify proteins, and two domain independent keypoint detectors. We conduct a case study with images from a publically available data set which are synthetically distorted using thin plate splines. The performance is assessed by the repeatability score, the probability of an image structure to be detected in original and distorted images with reasonable localization accuracy.

Image registration is an important ingredient in a wide variety of computer vision applications. Over the years countless algorithms emerged that allow for robust registration of image sequences. Unfortunately, high quality results still cannot be guaranteed in any case. Especially in interactive online systems that strongly rely on results of unsupervised registration algorithms, techniques for automatic quality analysis and failure compensation are indispensable. In this paper we present a new concept for an integrated and fully automatic detection and analysis of errors in registration. Based on a new metric for registration quality assessment, image differences are robustly detected. In addition, a hierarchical analysis scheme is proposed that allows distinguishing between various underlying error sources, all having different impacts on a registration result and requesting for individual compensation strategies.

Robots that support humans in their private homes by carrying out tasks of everyday life require sophisticated skills. Among others they have to provide multi-modal communication facilities, and also generic recognition and learning capabilities are indispensable. However, several of these skills, e.g., the recognition of objects, cannot be seen in isolation. Rather cognitive motivated approaches are on demand that integrate aspects of perception, reasoning and learning. Since all of these processes strongly rely on memorising and recalling data a visual memory component yields a fundamental building block of cognitive motivated system architectures. Such a memory usually subsumes different levels of abstraction ranging from low-level iconic to high-level categorial data and will often be organised hierarchically. Within this paper we propose an extended iconic scene representation for low level visual data, the SMARD memory module. It is particularly designed to meet the requirements of interactive and mobile robotic systems. These systems significantly benefit from such a memory component, however, at the same time they enforce strict constraints on its architecture, e.g., regarding flow of data and control or resource management strategies. Our approach is based on mosaic images that are computed following an incremental strategy. Due to polytopial coordinate frames they support easy data access and a direct analysis. The practical relevance of our approach is outlined by two example applications that illustrate its benefits with regard to intuitive and efficient human-robot interaction.

Automatic quality assessment of image registration results is an important issue in image processing. Many applications strongly depend on accurate registration results, sometimes even requiring automatic self-recovery from registration failures. In doing so it is not sufficient just to detect registration errors. Also a distinction between different error sources is necessary, as each source has an individual impact on the final registration result and requires specific compensation strategies. We present a new approach to automatically identify lens distortions in image pairs, known to have a significant impact on registration. The key idea is to analyse registration residuals and to learn a model of spatial residual distributions typical for distorted images. Our approach relies on a new metric for registration quality assessment and implements a regression scheme based on SVMs for predicting distortions in unknown data. The potential of the approach is demonstrated by experimental results on synthetic as well as real image data.

Image registration is an important ingredient in a wide variety of computer vision applications. Over the years countless algorithms emerged that allow for robust registration of image sequences. Unfortunately, high quality results still cannot be guaranteed in any case. Especially in interactive online systems that strongly rely on results of unsupervised registration algorithms, techniques for automatic quality analysis and failure compensation are indispensable. In this paper we present a new concept for an integrated and fully automatic detection and analysis of errors in registration. Based on a new metric for registration quality assessment, image differences are robustly detected. In addition, a hierarchical analysis scheme is proposed that allows to distinguish between various underlying error sources, all having different impacts on a registration result and requesting for individual compensation strategies.

Geometric registration of visual images is a fundamental intermediate processing step in a wide variety of computer vision applications that deal with image sequence analysis. 2D motion recovery and mosaicing, 3D scene reconstruction and also motion detection approaches strongly rely on accurate registration results. However, automatically assessing the overall quality of a registration is a challenging task. In particular, optimization criteria used in registration are not necessarily closely linked to the final quality of the result and often show a lack of local sensitivity. In this paper we present a new approach for an objective quality metric in 2D image registration. The proposed method is based on local structure analysis and facilitates voting-techniques for error pooling, leading to an objective measure that correlates well with the visual appearance of registered images. Since observed differences are furthermore classified in more detail according to various underlying error sources, the new measure not only yields a suitable base for objective quality assessment, but also opens perspectives towards an automatic and optimally adjusted correction of errors.

Geometric image registration by estimating homographies is an important processing step in a wide variety of computer vision applications. The 2D registration of two images does not require an explicit reconstruction of intrinsic or extrinsic camera parameters. However, correcting images for non-linear lens distortions is highly recommended. Unfortunately, standard calibration techniques are sometimes difficult to apply and reliable estimations of lens distortions can only rarely be obtained. In this paper we present a new technique for automatically detecting and categorising lens distortions in pairs of images by analysing registration results. The approach is based on a new metric for registration quality assessment and facilitates a PCA-based statistical model for classifying distortion effects. In doing so the overall importance for lens calibration and image corrections can be checked, and a measure for the efficiency of accordant correction steps is given.

One basic capability of interactive and mobile systems to cope with unknown situations and environments is active, sequence-based visual scene analysis. Image sequences provide static as well as dynamic and also 2D as well as 3D information about a certain scene. However, at the same time they require efficient mechanisms to handle their large data volumes. In this paper we introduce a new concept of a visual scene memory for interactive mobile systems that supports these systems with a space- and time-efficient data structure for representing iconic information. The memory is based on a new kind of mosaic images called multi-mosaics and allows to efficiently store and process sequences of stationary rotating and zooming cameras. Its main key features are polytopial reference coordinate frames and an online data processing strategy. The polytopes provide euclidean coordinates and thus allow the application of standard image analysis algorithms directly to the data yielding easy access and analysis, while online data processing preserves system interactivity. Additionally, mechanisms are included to properly handle multi-resolution data and to deal with dynamic scenes. The concept has been implemented in terms of an integrated system that can easily be included as an additional module in the architecture of interactive and mobile systems. As one prototypical example for possible fields of application the integration of the memory into the architecture of an interactive multi-modal robot is discussed emphasizing the practical relevancy of the new concept.

Natural human-robot interaction (HRI) is a key feature of mobile robot companions collaborating with humans. To achieve natural HRI, multiple communication modalities like vision, speech, and gestures have to be utilized. Besides, capabilities to emulate cognitive processes, e.g., object learning and object recognition, are essential. In this work we present a new approach to interactive object learning enabling multi-view object representation. To overcome a robot's limitation of having only one view point, we make use of an iconic memory consisting of previously acquired images. As the relevant scene area is unknown during construction of the iconic memory, a representation in the form of mosaic images is applied. The relevant image patches describing an object referenced by the user are selected through an object attention mechanism. The resulting multi-view object representations improve the flexibility of our interactive approach for object learning.

Processing visual data is an important ability of interactive systems to act in dynamically changing environments. Looking at the human visual and cognitive system this requires efficient mechanisms for data processing and storage as well as intelligent strategies for data acquisition. In this paper we present a visual memory supporting efficient representation of image sequences of active cameras in an online fashion. The memory is based on mosaic images extending the field of view of a camera in space and time. As one prototypical field of application for the memory active scene exploration is discussed. The temporally and spatially integrated data of the memory combined with additional feature maps serves as an ideal starting point for the selection of focus points. Hence the memory demonstrates the combination of efficient acquisition and storage of visual data and helps to provide interactive systems with high flexibility to operate in dynamically changing environments.

Die Dissertation stellt das neue Konzept der so genannten Multi-Mosaikbilder zur effizienten Repräsentation von Bildsequenzen aktiver Kameras vor. Das Konzept zielt darauf ab, interaktiven Systemen ein visuelles Gedächtnis für ikonische Daten zur Verfügung zu stellen, das eine flexiblere Nutzung visueller Informationen erlaubt und damit zu einer Erhöhung der Leistungsfähigkeit und einer Verbesserung der Kommunikationsmöglichkeiten interaktiver Systeme beitragen kann. Mosaikbilder sind ein Mechanismus, der eine kompakte Darstellung von Bildfolgen ermöglicht. Dabei werden alle Bilder einer Folge unter Eliminierung redundanter Anteile und einer signifikanten Reduktion des Datenvolumens zu einem Bild verschmolzen. Der Einsatz existierender Verfahren zur Mosaikbildberechnung in interaktiven Systemen ist allerdings häufig aufgrund der verwendeten Koordinatensysteme und der gewählten Verarbeitungsstrategien mit Schwierigkeiten verbunden. Das neue Konzept der Multi-Mosaikbilder erweitert und ergänzt daher gängige Ansätze, um einen Einsatz dieser Techniken auch in ressourcenbeschränkten, interaktiven Systemen zu ermöglichen. Das neue Konzept gründet im Kern auf der Verwendung polyedrischer Koordinaten, in denen Bilddaten stationärer rotierender Kameras weitgehend verzerrungsfrei repräsentiert werden können. Gleichzeitig wird die direkte Anwendung existierender Bildverarbeitungsalgorithmen auf die Bilddaten ermöglicht. Dies gewährleistet einen einfachen Zugriff auf die gespeicherten Informationen und erlaubt interaktiven Systemen einen flexiblen Umgang mit der Datenstruktur. Darüber hinaus unterstützt das neue Konzept eine inkrementelle Online-Berechnung der Mosaikbilder, die Verarbeitung von Bildfolgen mit statischen und dynamischen Inhalten, und die adäquate Darstellung von Bilddaten mit verschiedenen Zoomstufen innerhalb einer Auflösungshierarchie. Die praktische Relevanz des entwickelten Konzeptes wurde durch die Realisierung einer aktiven Szenenexploration auf Basis der Multi-Mosaikbilder belegt, die den Kreislauf aus Datenrepräsentation und aktiver Akquisition schließt. Außerdem erfolgte eine prototypische Integration des visuellen Speichers in die Architektur des interaktiven, mobilen Roboters BIRON, dessen Fähigkeiten zum Lernen und Wiedererkennen von Objekten auf diese Weise verbessert werden sollen.

Visual data acquired with active cameras yields an important source of information for interactive systems. However, since image sequences usually comprise large data volumes and notable portions of redundant information analysis is often diffcult. Hence, data structures are required that allow for compact representation of image sequences. In this paper we introduce our concept of a visual scene memory. The memory is based on mosaic images enabling compact image sequence representation by fusing all sequence images into one single frame while eliminating redundancies. Since interactive systems put special demands on mosaicing techniques we developed a new mosaic concept called multi-mosaics well-suited to be used with interactive systems. The memory is focussed on adaquate representation of iconic data, however, not restricted to it. Rather higher-level data, particularly motion data as well as data suitable for active camera control are additionally included completing the visual scene representation.

Interactive systems using active cameras produce a large amount of image data to be processed. The task of a visual memory is to represent this data in an efficient way and thus assisting subsequent image analysis algorithms. In this paper we extent our mosaic based approach to large scenes where distortions occur if the camera rotation covers a wide angle. This problem can be solved by using a spherical mosaic surface. However, since most vision algorithms are created for planar images we propose to tile the sphere with planar patches resulting in a mosaic polytope we call multi-mosaic. A more spacious scene can be represented by a number of such multi-mosaics taken from different spatial positions. If a mobile robot is used for image acquisition the path in between can additionally be represented by manifold mosaics to connect the positions.

Mosaicing is a technique to efficiently condense the static information of an image sequence within one extended mosaic image. The core of mosaicing is to estimate a global transformation between images due to the global camera motion. This is usually accomplished by either matching segmented image features or exploiting all iconic image data directly within a featureless approach. In this paper we propose to combine aspects from both techniques where we abandon to segment features, however select pixels to be used for parameter estimation based on structural image data and information about independently moving scene parts. While this results in a speed up of the estimation process the main focus is to improve robustness with respect to ambiguities arising from homogeneous image regions and to motion in the scene.

Interactive systems using active cameras produce a big amount of image data to be processed. The task of a visual memory is to represent this data in an efficient way and thus assisting subsequent image analysis algorithms. In this paper we extent our mosaic based approach to large scenes where distortions occure if the camera rotation covers a wide angle. This problem can be solved by using a spherical mosaic surface. However, since most vision algorithms are created for planar images we propose to tile the sphere with planar patches resulting in a mosaic polytope we call multi-mosaic. A more spacious scene can be represented by a number of such multi-mosaics taken from different spatial positions. If a mobile robot is used for image acquisition the path in between can additionally be represented (by manifold mosaics) to connect the positions.

Exploring a scene with an active camera yields an image sequence containing implicit information of the environment, but also a considerable amount of redundancies. In this paper the system AViSMo is described to explicitly represent the information about the scene in a visual memory. It is based on the construction of a mosaic image containing non-redundantly iconic data of static scene parts. This mosaic is augmented with trajectories of moving objects, activities detected from image data in a bottom-up fashion, and external high-level data. AViSMo is a highly modular and extendable system, capable of communicating with external components using a variety of transport mechanisms. As an example of such an external component it comprises a GUI to control and visualize the mosaicing process. Thus, it is ideally suited to supply interactive systems with a visual scene memory.

One important source of information in scene understanding is given by actions performed either by human actors or robots. In this paper an approach to recognition and low-level interpretation of actions is presented. Since actions are characterized by specific motion patterns of moving objects, recognition is done by detecting such motion patterns as specific constellations of interactions between moving objects. First of all, motion detection and tracking algorithms are applied to extract correspondences between moving objects in consecutive images of a sequence. Subsequently these are represented with a graph data-structure for further analysis. To detect interactions of moving objects robustly a short history of motion of objects is traced using a finite-state automaton. Finally activities are segmented based on detected interactions. Since robust motion data are required consistency checks and corrections of the acquired motion data are performed in parallel.

Mosaic images provide an efficient representation of image sequences and simplify scene exploration and analysis. However, the application of conventional methods to generate mosaics of scenes with moving objects causes integration errors and a loss of dynamic information. In this paper a method to compute mosaics of dynamic scenes is presented addressing the above mentioned problems. Moving pixels are detected in the images and not integrated in the mosaic yielding a consistent representation of the static scene background. Furthermore, dynamic object information is extracted by tracking moving regions. To account for unavoidable variances in region segmentation topologically neighboring regions are grouped into sets before tracking. The regions' and objects' motion characteristics are described by trajectories. Along with the background mosaic they provide a complete representation of the underlying scene which is idealy suited for further analysis.