Robust face recovery for hybrid surface visualization
Carlos Andújar, Pere Brunet, J. Esteve, Eva Monclús, Isabel Navazo, Àlvar Vinacua · Vision Modeling and Visualization · 2003
Abstract Modern adquisition techniques provide 3D datasetswith unprecedented detail. Both triangle-based andpoint-based techniques have been proposed for effi-cient rendering of these objects. We present a newtechniqueforinteractivevisualizationofvolumetricdata which combines polygonal faces for flat areasand oriented points for curved regions. The paperfocuses on extracting large polygonal faces fromvoxel data. The key ingredient is a region-growingalgorithmwhichmaintainsa discreterepresentationof the set of planes that satisfy a separability prob-lem. A robust representation of this set of planesbased on a discretized 4D convex polytope is alsodiscussed. The proposed face extraction algorithmisabletorecoverinformationatsub-voxelprecisionand automatically detects sharp edges. 1 Introduction Current adquisition techniques such as laser rangescanners and medical imaging devices provide verylarge 3D datasets with unprecedented detail. Overthelastfewyears,anumberoftechniqueshavebeenproposed for the interactive visualization of suchlarge datasets.Inaconventional approach,theseobjectsareren-dered through a triangle mesh extracted from theraw data. Most surface reconstruction techniquesfromdatasampledon3Dgridsareextensionsofthewell-known Marching Cubes (MC) algorithm [12].These techniques often suffer from two problems:uniform over-tessellation and alias artifacts. Strate-gies for alleviating over-tessellation include Dis-creteMC[13]andmesh simplification; ontheotherhand, several techniques address feature-preservingextraction [10, 8].In recent years, strategies for discretizing syn-thetic graphics objects into point-based representa-tions [18, 15] have become an attractive alternativefor rendering very large scanned objects. Compar-ing both approaches, point primitives outperformtrianglesincurvedareaswhereastriangleprimitivescould be a more compact representation for flat re-gions, provided that these regions are convenientlydetected. Therefore, for a method combining bothkinds of primitives, efficient detection of flat re-gionsandfeature-preservingreconstructionoflargepolygonal faces is required.This paper focuses on the problem of extractinglarge polygonal faces from volume datasets. Weshow that resulting meshes can be used in combi-nation with point-rendering approaches for efficientrendering of complex datasets involving a mixtureof planar and curved regions.The input model is first converted into a setof face-connected voxels called