Breast cancer continues to be the most common type of cancer among women. Early detection of breast cancer is key to effective treatment. The presence of clusters of fine, granular microcalcifications in mammographic images can be a primary sign of breast cancer. The malignancy of any cluster of microcalcification cannot be reliably determined by radiologists from mammographic images and need to be assessed through histology images. In this paper, a novel method of mammographic microcalcification classification is described using the local topological structure of microcalcifications. Unlike the statistical and texture features of microcalcifications, the proposed method focuses on the number of microcalcifications in local clusters, the distance between them, and the number of clusters. The initial evaluation on the Digital Database for Screening Mammography (DDSM) database shows promising results with 86% accuracy and findings which are in line with clinical perception of benign and malignant morphological appearance of microcalcification clusters.

Detecting and tracking people in a meeting room is very important for many applications. In order to detect people in a meeting room with no prior knowledge (e.g. background model) and regardless of whether their motion is slow or significant, this paper proposes a coarse-to-fine people detection algorithm by combining a novel motion detection process, namely, adaptive accumulated frame differencing (AAFD) combined with corner features. Firstly, the region of movement is extracted adaptively using AAFD, then motion corner features are extracted. Finally, the minimum area rectangle fitting these corners is found. The proposed algorithm is evaluated using the AMI meeting data set and this indicates promising results for people detection.

Groupwise image alignment automatically provides non-rigid registration across a set of images. It has found applications in facial image analysis and medical image analysis by automatically generating statistical models of shape and appearance. The main approaches used previously include iterative and graph-based approaches. In iterative approaches, the registration of each image is iteratively updated to minimise an error measure across the set. Various metrics and optimisation strategies have been proposed to achieve this. Graph-based methods perform registration of each pair of images in the set, to form a weighted graph of the ''distance'' between all the images, and then finds the optimal paths between the most central image and every other image. In this paper, we use a graph-based approach to perform initialisation, which is then refined with an iterative approach. Pairwise registration is performed using demons registration, then shortest paths identified in the resulting graph are used to provide an initial warp for each image by concatenating warps along the path. The warps are refined using an iterative Levenberg-Marquardt minimisation to the mean, based on updating the locations of a small number of points and incorporating a stiffness constraint. This optimisation approach is efficient, has very few free parameters to tune and we show how to tune the few remaining parameters. We compare the combined approach to both the iterative and graph-based approaches used independently. Results demonstrate that the combined method improves the alignment of various datasets, including two face datasets and a difficult medical dataset of prostate MRI images.

In Full-Reference Image Quality Assessment (FR-IQA) images are compared with ground truth images that are known to be of high visual quality. These metrics are utilized in order to rank algorithms under test on their image quality performance. Throughout the progress of Monte Carlo rendering processes we often wish to determine whether images being rendered are of sufficient visual quality, without the availability of a ground truth image. In such cases FR-IQA metrics are not applicable and we instead must utilise No-Reference Image Quality Assessment (NR-IQA) measures to make predictions about the perceived quality of unconverged images. In this work we propose a deep learning approach to NR-IQA, trained specifically on noise from Monte Carlo rendering processes, which significantly outperforms existing NR-IQA methods and can produce quality predictions consistent with FR-IQA measures that have access to ground truth images.

The vision of fully simulating characters and their environments has the potential to offer rich interactions between characters and objects in the virtual world. However, this introduces a challenging problem similar to controlling robotic figures: computing the necessary torques to perform a given task. In this paper, we address the problem of transferring hand-crafted kinematic motions to a fully simulated figure, by computing open-loop controls necessary to reproduce the target motion. One key ingredient to successful control is the mechanical feasibility of the target motion. While several methods have been successful at replicating human captured motion, there has not yet been a method capable of handling the case of artist-authored key-framed movements that can violate the laws of physics or go beyond the mechanical limits of the character. Due to the curse of dimensionality, sampling-based optimization methods typically restrict the search to a narrow band which limits exploration of feasible motions—resulting in a failure to reproduce the desired motion when a large deviation is required. In this paper, we solve this problem by combining a window-based breakdown of the controls on the temporal dimension, together with a global wide search strategy that keeps locally sub-optimal samples throughout the optimization.

We introduce Shadow Retargeting which maps real shadow appearance to virtual shadows given a corresponding deformation of scene geometry, such that appearance is seamlessly maintained. By performing virtual shadow reconstruction from un-occluded real shadow samples observed in the camera frame, we recover the deformed shadow appearance efficiently. Our method uses geometry priors for the shadow casting object and a planar receiver surface. Inspired by image retargeting approaches [VTP10] we describe a novel local search strategy, steered by importance based deformed shadow estimation. Results are presented on a range of objects, deformations and illumination conditions in real-time Augmented Reality (AR) on a mobile device. We demonstrate the practical application of the method in generating otherwise laborious in-betweening frames for 3D printed stop motion animation.

Abstract In this paper, a novel solution is provided for the sticky lip problem in computer facial animation, recreating the way the lips stick together when drawn apart in speech or in the formation of facial expressions. Traditional approaches to modelling this rely on an artist estimating the correct behaviour. In contrast, this paper presents a physically-based model. The mouth is modelled using the total Lagrangian explicit dynamics finite element method, with a new breaking element modelling the saliva between the lips. With this approach, subtle yet complex behaviours are recreated implicitly, giving rise to more realistic movements of the lips. The model is capable of reproducing varying degrees of stickiness between the lips, as well as asymmetric effects.

Analyses of large 3D particle datasets typically involve many different exploration and visualization steps. Interactive exploration techniques are essential to reveal and select interesting subsets like clusters or other sophisticated structures. State-of-the-art techniques allow for context-aware selections that can be refined dynamically. However, these techniques require large amounts of memory and have high computational complexity which heavily limits their applicability to large datasets. We propose a novel, massively parallel particle selection method that is easy to implement and has a processing complexity of O(n*k) (where n is the number of particles and k the maximum number of neighbors per particle) and requires only O(n) memory. Furthermore, our algorithm is designed for GPUs and performs a selection step in several milliseconds while still being able to achieve high-quality results.

With four percent of the working population employed in call centers in both the United States and the UK, the contact center industry represents a sizable proportion of modern industrial landscapes. As with most modern industries, data collection is de rigueur, producing gigabytes of call records that require analysis. The scatterplot is a well established and understood form of data visualization dating back to the 17th century. In this paper we present an application for visualizing large call centre data sets using hardware-accelerated scatterplots. The application utilizes a commodity graphics card to enable visualization of a month's worth of data, enabling fast filtering of multiple attributes. Filtering is implemented using the Open Computing Language (OpenCL), providing significant performance improvement over traditional methods. We demonstrate the value of our application for exploration and analysis of millions of call events from a real-world industry partner. Domain expert feedback from our industrial partners is reported.

Developing a positive relationship between a business and its customers is vital to success. The outcome of any customer interaction can determine future patronage of the business. Many industry's only point of interaction with their customers is through a contact centre where everything from sales to complaints are handled. This places tremendous importance on the operational efficiency of the contact centre and the level of care provided to the customers. These customer interactions are recorded and archived in large databases, but undertaking insightful analysis is challenging due to both the size and complexity of the data. We present a visual solution to the tracking of customer interactions at a large scale. RiverState visualises the collective flow of callers through the process of interacting with a contact centre using a river metaphor. We use finite state transition machines with customised edges to depict millions of events and the states callers go through to complete their journey. We implement a range of novel features to enhance the analytical qualities of the application, and collect feedback from domain experts to analyse and evaluate the use of the software.

Graphic design and visualization are becoming fundamental components of education. The use of advanced visual design in pedagogy is growing and evolving rapidly. One of their aims is to enhance the educational process by facilitating better understanding of the subject with the use of graphical representation methods. Research papers in this field offer important opportunities to examine previously completed experiments and extract useful educational outcomes. This paper analyzes and classifies pedagogical visualization research papers to increase understanding in this area. To our knowledge, this is the first (work-in-progress) survey paper on advanced visualization for education. We categorize related research papers into original subject groups that enable researchers to compare related literature. Our novel classification enables researchers to find both mature and unexplored directions which can inform directions for future work. This paper serves as a valuable resource for both beginners and experienced researchers who are interested in interactive visualization for education.

Deep neural networks have been successfully applied to problems such as image segmentation, image super-resolution, coloration and image inpainting. In this work we propose the use of convolutional neural networks (CNN) for image inpainting of large regions in high-resolution textures. Due to limited computational resources processing high-resolution images with neural networks is still an open problem. Existing methods separate inpainting of global structure and the transfer of details, which leads to blurry results and loss of global coherence in the detail transfer step. Based on advances in texture synthesis using CNNs we propose patch-based image inpainting by a CNN that is able to optimize for global as well as detail texture statistics. Our method is capable of filling large inpainting regions, oftentimes exceeding the quality of comparable methods for high-resolution images. For reference patch look-up we propose to use the same summary statistics that are used in the inpainting process.

This paper addresses the automatic segmentation of teeth in volumetric Computed Tomography (CT) scans of the human skull. Our approach is based on a convolutional neural network employing 3D volumetric convolutions. To tackle data scale issues, we apply a hierarchical coarse-to fine approach combining two CNNs, one for low-resolution detection and one for highresolution refinement. In quantitative experiments on 40 CT scans with manually acquired ground truth, we demonstrate that our approach displays remarkable robustness across different patients and device vendors. Furthermore, our hierarchical extension outperforms a single-scale segmentation, and network size can be reduced compared to previous architectures without loss of accuracy.

Voronoi tree maps are an important milestone in information visualization, representing a substantial advancement of the original tree maps concept. We address a less-studied variant of Voronoi tree maps that uses multiplicative-weighted Voronoi diagrams. We highlight the merits of this variant, and discuss the difficulties that might have discouraged further exploration, proposing insights for overcoming these difficulties.

In this paper we propose and analyse a method for watermarking 3D printed objects, concentrating on the watermark retrieval problem. The method embeds the watermark in a planar region of the 3D printed object in the form of small semi-spherical or cubic bumps arranged at the nodes of a regular grid. The watermark is extracted from a single image of the watermarked planar region through a Convolutional Neural Network. Experiments with 3D printed objects, produced by filaments of various colours, show that in most cases the retrieval method has a high accuracy rate.

In this study, we combine computer vision and visualisation/data exploration to analyse magnetic resonance imaging (MRI) data and detect garden peas inside the stomach. It is a preliminary objective of a larger project that aims to understand the kinetics of gastric emptying. We propose to perform the image analysis task as a multi-objective optimisation. A set of 7 equally important objectives are proposed to characterise peas. We rely on a cooperation co-evolution algorithm called 'Fly Algorithm' implemented using NSGA-II. The Fly Algorithm is a specific case of the 'Parisian Approach' where the solution of an optimisation problem is represented as a set of individuals (e.g. the whole population) instead of a single individual (the best one) as in typical evolutionary algorithms (EAs). NSGA-II is a popular EA used to solve multi-objective optimisation problems. The output of the optimisation is a succession of datasets that progressively approximate the Pareto front, which needs to be understood and explored by the end-user. Using interactive Information Visualisation (InfoVis) and clustering techniques, peas are then semi-automatically segmented.

Cartograms are a popular and useful technique for depicting geo-spatial data. Dorling style and rectangular cartograms are very good for facilitating comparisons between unit areas. Each unit area is represented by the same shape such as a circle or rectangle, and the uniformity in shapes facilitates comparative judgment. However, the layout of these more abstract shapes may also simultaneously reduce the map's legibility and increase error. When we integrate univariate data into a cartogram, the recognizability of cartogram may be reduced. There is a trade-off between information recognition and geo-information accuracy. This is the inspiration behind the work we present. We attempt to increase the map's recognizability and reduce error by introducing topological features into the cartographic map. Our goal is to include topological features such as a river in a Dorling-style or rectangular cartogram to make the visual layout more recognizable, increase map cognition and reduce geospatial error. We believe that compared to the standard Dorling and rectangular style cartogram, adding topological features provides familiar geo-spatial cues and flexibility to enhance the recognizability of a cartogram.

The use of colour mapping is fundamental to visualisation research. It acts as an additional layer beyond rendering in the spatial dimensions and provides a link between values in any dataset. When designing and building visualisation research software, the process of creating and managing a colour mapping system can be time-consuming and complex. Existing alternatives offer niche features and require complex dependencies or installations. We present Spectrum; an open source colour map management library that is developer friendly with no installation required, and that offers a wide variety of features for the majority of use cases. We demonstrate the utility of the library through simple snippets of code and a number of examples which illustrate its ease of use and functionality, as well as a video demonstrating the installation and use of the library in under two minutes. It is a very valuable jump-start tool for developers and researchers who need to focus on other tasks.

Text visualization is a rapidly growing sub-field of information visualization and visual analytics. There are many approaches and techniques introduced every year to address a wide range of tasks and enable researchers from different disciplines to obtain leading-edge knowledge from digitized collections. This can be challenging particularly when the data is massive. Additionally, the sources of digital text have spread substantially in the last decades in various forms, such as web pages, blogs, twitter, email, electronic publications, and books. In response to the explosion of text visualization research literature, the first survey article was published in 2010. Furthermore, there are a growing number of surveys that review existing techniques and classify them based on text research methodology. In this work, we aim to present the first Survey of Surveys (SoS) that review all of the survey and state-of-the-art papers on text visualization techniques and provide an SoS classification. We study and compare the surveys, and categorize them into 5 groups: (1) document-centered, (2) user task analysis, (3) cross-disciplinary, (4) multifaceted, and (5) satellite-themed. We provide survey recommendations for researchers in the field of text visualization. The result is a very unique, valuable starting point and overview of the current state-of-the-art in text visualization research literature.

3D models of transportation objects like a road, bridge, underpass, etc. are required in many domains including military training, land development, etc. While remote sensed images and LiDaR data can be used to create approximate 3D representations, detailed 3D representations are difficult to create automatically. Instead, interactive tools are used with rather laborious effort. For example, the top commercial interactive model generator we tried required 94 parameters in all for different bridge types. In this paper, we take a different path.We automatically derive these parameter values from GIS (Geographic Information Systems) data, which normally contains detailed information of these objects, but often only implicitly. The framework presented here transforms GIS data into a knowledge base consisting of assertions. Spatial/numeric relations are handled through plug-ins called property extractors whose results get added to the knowledge base, used by a reasoning engine to infer object properties. A number of properties have to be extracted from images, and are dependent on the accuracy of computer vision methods. While a comprehensive property extractor mechanism is work in progress, . a prototype implementation illustrates our framework for bridges with GIS data from the real world. To the best of our knowledge, our framework is the first to integrate knowledge inference and uncertainty for extracting landscape object properties by fusing facts from multi-modal GIS data sources.

Choropleth maps are an invaluable visualization type for mapping geo-spatial data. One advantage to a choropleth map over other geospatial visualizations such as cartograms is the familiarity of a non-distorted landmass. However, this causes challenges when an area becomes too small in order to accurately perceive the underlying color. When does size matter in a choropleth map? We conduct an experiment to verify the relationship between choropleth maps, their underlying color map, and a user's perceivability. We do this by testing a user's perception of color relative to an administrative area's size within a choropleth map, as well as user-preference of fixed-locale maps with enforced minimum areas. Based on this initial experiment we can make the first recommendations with respect to a unit area's minimum size in order to be perceivably useful.

This paper presents a multi-disciplinary overview of research evaluating virtual reality (VR). The main aim is to review and classify VR research based on several metrics: presence and immersion, navigation and interaction, knowledge improvement, performance and usability. With the continuous development and consumerisation of VR, several application domains have studied the impact of VR as an enhanced alternative environment for performing tasks. However, VR experiment results often cannot be generalised but require specific datasets and tasks suited to each domain. This review and classification of VR metrics presents an alternative metrics-based view of VR experiments and research.