Enrico Gobbetti

ViVa: the virtual vascular project

The aim of the virtual vascular project (ViVa) is to develop tools for the modern hemodynamicist and cardiovascular surgeon to study and interpret the constantly increasing amount of information being produced by noninvasive imaging equipment. In particular, we are developing a system able to process and visualize three dimensional (3D) medical data, reconstruct the geometry of arteries of specific patients, and simulate blood flow in them. The initial applications of the system will be for clinical research and training purposes. In a later stage, we explore the application of the system to surgical planning. ViVa is based on an integrated set of tools, each dedicated to a specific aspect of the data processing and simulation pipeline: image processing and segmentation; real time 3D volume visualization; 3D geometry reconstruction; 3D mesh generation; and blood flow simulation and visualization.

Light calibration and quality assessment methods for Reflectance Transformation Imaging applied to artworks' analysis

In this paper we analyze some problems related to the acquisition of multiple illumination images for Polynomial Texture Maps (PTM) or generic Reflectance Transform Imaging (RTI). We show that intensity and directionality nonuniformity can be a relevant issue when acquiring manual sets of images with the standard highlight-based setup both using a flash lamp and a LED light. To maintain a cheap and flexible acquisition setup that can be used on field and by non-experienced users we propose to use a dynamic calibration and correction of the lights based on multiple intensity and direction estimation around the imaged object during the acquisition. Preliminary tests on the results obtained have been performed by acquiring a specifically designed 3D printed pattern in order to see the accuracy of the acquisition obtained both for spatial discrimination of small structures and normal estimation, and on samples of different types of paper in order to evaluate material discrimination. We plan to design and build from our analysis and from the tools developed and under development a set of novel procedures and guidelines that can be used to turn the cheap and common RTI acquisition setup from a simple way to enrich object visualization into a powerful method for extracting quantitative characterization both of surface geometry and of reflective properties of different materials. These results could have relevant applications in the Cultural Heritage domain, in order to recognize different materials used in paintings or investigate the ageing status of artifacts’ surface.

Digital Mont'e Prama: 3D Cultural Heritage presentations in museums and anywhere.

We present an interactive visualization system developed for the valorization of an extraordinary collection of protostoric Mediterranean sculptures, which depict models of buildings (cone-shaped stone towers), as well as larger-than-life human figures. The architecture is based on scalable components for efficient distribution and adaptive rendering of extremely detailed surface meshes, as well as a simple and effective interactive camera controller tailored for touch interaction. The user interface has been designed for targeting both small screens and large display systems, and in a way that casual users can easily and naturally explore the models with fast learning curves. Furthermore, a thumbnail-based point-of-interest selector enable users to explore 3D views with information presented as 2D overlays decorating the 3D scene. The system components have been integrated in different interactive applications, ranging from large-screen museal setups and low end mobile devices both with very high visual quality. The capabilities of the museal systems have been demonstrated in a variety of temporal and permanent exhibitions, where they have been extensively used by tens of thousands of visitors.

Split-voxel: a simple discontinuity-preserving voxel representation for volume rendering

The most common representation of volumetric models is a regular grid of cubical voxels with one value each, from which a smooth scalar field is reconstructed. However, common real-world situations include cases in which volumes represent physical objects with well defined boundaries separating different materials, giving rise to models with quasi-impulsive gradient fields. In our split-voxel representation, we replace blocks of N3 voxels by one single voxel that is split by a feature plane into two regions with constant values. This representation has little overhead over storing precomputed gradients, and has the advantage that feature planes provide minimal geometric information about the underlying volume regions that can be effectively exploited for volume rendering. We show how to convert a standard mono-resolution representation into a out-of-core multiresolution structure, both for labeled and continuous scalar volumes. We also show how to interactively explore the models using a multiresolution GPU raycasting framework. The technique supports real-time transfer function manipulation and proves particularly useful for fast multiresolution rendering, since accurate silhouettes are preserved even at very coarse levels of detail.

Automated color clustering for medieval manuscript analysis

Given a color image of a medieval manuscript page, we propose a simple, yet efficient algorithm for automatically estimating the number of its color-based pixel groups, K. We formulate this estimation as a minimization problem, where the objective function assesses the quality of a candidate clustering. Rather than using all the features of the given image, we carefully select a subset of features to perform clustering. The proposed algorithm was extensively evaluated on a dataset of 2198 images (1099 original images and their 1099 variants produced by modifying both spatial and spectral resolutions of the originals) from the Yales Institute for the Preservation of Cultural Heritage (IPCH). The experimental results show that it is able to yield satisfactory estimates of K for these test images.

An Automatic Word-spotting Framework for Medieval Manuscripts

We present a completely automatic and scalable framework to perform query-by-example word-spotting on medieval manuscripts. Our system does not require any human intervention to produce a large amount of annotated training data, and it provides Computer Vision researchers and Cultural Heritage practitioners with a compact and efficient system for document analysis. We have executed the pipeline both in a single-manuscript and a cross-manuscript setup, and we have tested it on a heterogeneous set of medieval manuscripts, that includes a variety of writing styles, languages, image resolutions, levels of conservation, noise and amount of illumination and ornamentation. We also present a precision/recall based analysis to quantitatively assess the quality of the proposed algorithm.

Automatic geometric calibration of projector-based light-field displays

We present a novel calibration method for continuous multiview (light field) projection-based displays using a single uncalibrated camera and four fiducial markers. Calibration starts from a simple parametric description of the display layout. First, individual projectors are calibrated through parametric optimization of an idealized pinhole model. Then, the overall display and projector parameterization is globally optimized. Finally, independently for each projector, remaining errors are corrected through a rational 2D warping function. The final parameters are available to rendering engines to quickly compute forward and backward projections. The technique is demonstrated in the calibration of a large-scale horizontal-parallax-only 35MPixels light field display.

The V-City Project

3D geoinformatics have entered the digital age, hesitantly in some areas, and rampantly in others. Google Earth and Microsoft Virtual Earth are household names. However, these projects are limited to textured 3D landscapes, aerial 2D images and a few boxy building envelopes. The V-City project is a European research initiative to surpass these limitations, and create a system for intuitively exploring large urban areas with a high degree of detail. Bringing together technologies from geoinformatics, virtual reality, computer graphics, and computer vision, the system constructs detailed 3D city models from geopositioned aerial images and building footprints. For networked browsing, city models are compressed and streamed for interactive viewing of entire landscapes. A unique tactile table has also been developed to let multiple users visualize the same city model in stereo 3D, and interact with it simultaneously using hand gestures.

Techniques for seamless color registration and mapping on dense 3D models

Today’s most widely used 3D digitization approach is a combination of active geometric sensing, mainly using laser scanning, with active or passive color sensing, mostly using digital photography. Producing a seamless colored object, starting from a geometric representation and a set of photographs, is a data fusion problem requiring effective solutions for image-to-geometry registration, and color mapping and blending. This chapter provides a brief survey of the state-of-the-art solutions, ranging from manual approaches to fully scalable automated methods.