The task of assigning geographic coordinates to textual resources plays an increasingly central role in geographic information retrieval. The ability to select those terms from a given collection that are most indicative of geographic location is of key importance in successfully addressing this task. However, this process of selecting spatially relevant terms is at present not well understood, and the majority of current systems are based on standard term selection techniques, such as $(\chi^2)$ or information gain, and thus fail to exploit the spatial nature of the domain. In this paper, we propose two classes of term selection techniques based on standard geostatistical methods. First, to implement the idea of spatial smoothing of term occurrences, we investigate the use of kernel density estimation (KDE) to model each term as a two-dimensional probability distribution over the surface of the Earth. The second class of term selection methods we consider is based on Ripley's K statistic, which measures the deviation of a point set from spatial homogeneity. We provide experimental results which compare these classes of methods against existing baseline techniques on the tasks of assigning coordinates to Flickr photos and to Wikipedia articles, revealing marked improvements in cases where only a relatively small number of terms can be selected.

Bas-relief is an art form part way between sculpture and drawing. In this paper, we present an algorithm for generating a bas-relief from a single image, inspired by the process that artists use to create reliefs. We do not aim to recover exact depth values for objects in the image, which is a tricky computer vision problem, requiring assumptions that are rarely satisfied. Instead, we determine layers based on relative depth ordering of objects (and their parts) in the image, and use this information to construct surfaces in the 3D relief model. Feature lines are extracted and used to build a new region-based representation of the input image. During surface construction, a base surface is first generated; it is then augmented using both intensity and gradient information from the original image. To prevent depth errors arising due to augmentation, a feedback process is used to refine the output. Our experimental results show the generated bas-reliefs have smooth boundaries with appropriate height relationships, a key property of bas-reliefs created by artists. We demonstrate that our algorithm works well for a range of input images, including human faces, flowers and animals.

This paper summarises work on low-discrepancy sampling of parametric surfaces and meshes based on use of space-filling curves, and suggests two applications of such work: shape retrieval based on distance histograms, and non-photorealistic stroke-based rendering.

We introduce a novel stratified sampling technique for mesh surfaces that gives the user control over sampling density and anisotropy via a tensor field. Our approach is based on sampling space-filling curves mapped onto mesh segments via parametrizations aligned with the tensor field. After a short preprocessing step, samples can be generated in real time. Along with visual examples, we provide rigorous spectral analysis and differential domain analysis of our sampling. The sample distributions are of high quality: they fulfil the blue noise criterion, so have minimal artifacts due to regularity of sampling patterns, and they accurately represent isotropic and anisotropic densities on the plane and on mesh surfaces. They also have low discrepancy, ensuring that the surface is evenly covered.

Centroidal Voronoi tessellations and optimal Delaunay triangulations can be approximated efficiently by non-linear optimisation algorithms. This paper demonstrates that the point distribution used to initialise the optimisation algorithms is important. Compared to conventional random initialisation, certain low-discrepancy point distributions help convergence towards more spatially regular results and require fewer iterations for planar and volumetric tessellations.

We present the results of our submission to the MediaEval 2012 Placing Task. We used a framework that combines language models and similarity search, which improves our system from last year by using a dierent feature selection technique, extending our similarity search, tapping into new types of information for videos without any tags and including the use of SIFT features.

The combination of mobile communication technology with location and orientation aware digital cameras has introduced increasing interest in the exploitation of 3D city models for applications such as augmented reality and automated image captioning. The effectiveness of such applications is, at present, severely limited by the often poor quality of semantic annotation of the 3D models. In this paper, we show how freely available sources of georeferenced Web 2.0 information can be used for automated enrichment of 3D city models. Point referenced names of prominent buildings and landmarks mined from Wikipedia articles and from the OpenStreetMaps digital map and Geonames gazetteer have been matched to the 2D ground plan geometry of a 3D city model. In order to address the ambiguities that arise in the associations between these sources and the city model, we present procedures to merge potentially related buildings and implement fuzzy matching between reference points and building polygons. An experimental evaluation demonstrates the effectiveness of the presented methods.

A novel point sampling framework for polygonal meshes is presented, based on sampling a mesh according to a density-controlled low-discrepancy distribution. The local sampling density can be controlled by a density func- tional defined by the user, e.g. to preserve local features, or to achieve desired data reduction rates. To sample the mesh, it is cut into a disc topology, and a parametrisation is generated. The parameterised mesh is sampled using a Hilbert curve in the parameter domain, which is adapted to parametric distortions and mapped onto the mesh. 1D sample points along the Hilbert curve are then generated, correcting for parametric distortion and a user- specified local density, to give a density-controlled low-discrepancy sampling of the mesh. After a pre-processing step, the sampling density can be adjusted in real-time. Experiments show that this approach can quickly resample existing meshes with low discrepancy samples. The effectiveness and speed of the approach are demonstrated by applying it to viewpoint dependent rendering, level of detail representation, and interactive remeshing.

Low-discrepancy point distributions exhibit excellent uniformity properties for sampling in applications such as rendering and measurement. We present an algorithm for generating low-discrepancy point distributions on arbitrary parametric surfaces using the idea of converting the 2D sampling problem into a 1D problem by adaptively mapping a space-filling curve onto the surface. The 1D distribution takes into account the parametric mapping by employing a corrective approach similar to histogram equalisation to ensure that it gives a 2D low-discrepancy point distribution on the surface. This also allows for control over the local density of the distribution, e.g. to place points more densely in regions of higher curvature. To allow for parametric distortion, the space-filling curve is generated adaptively to cover the surface evenly. Experiments show that this approach efficiently generates low-discrepancy distributions on arbitrary parametric surfaces and creates nearly as good results as well-known low-discrepancy sampling methods designed for particular surfaces like planes and spheres. However, we also show that machine-precision limitations may require surface reparameterisation in addition to adaptive sampling.