Service interruption on Monday 11 July from 12:30 to 13:00: all the sites of the CCSD (HAL, EpiSciences, SciencesConf, AureHAL) will be inaccessible (network hardware connection).
Skip to Main content Skip to Navigation

Cartes de saillance et évaluation de la qualité des maillages 3D

Anass Nouri 1 
1 Equipe Image - Laboratoire GREYC - UMR6072
GREYC - Groupe de Recherche en Informatique, Image et Instrumentation de Caen
Abstract : The glance of eachhuman being is attracted by specific areas into 3D objects (that can be represen- ted bymeshes). This attraction depends on the degree of saliency exposed by these areas. The first goal of this thesis is to propose an approach for detecting visual salient areas on 3D non-colored meshes. We consider that a vertex as salient if it strongly stands out from its local neighborhood and if its geometric configuration is different from its adjacent vertices. For this, we characte- rize the surface of the 3D mesh by the use of a local vertex descriptor in the form of an adaptive patch. This descriptor is used as a basis for similarity computation and integrated into a weighted multi-scale saliencycomputation.We propose also an extension of our visual saliencymodel to 3D colored meshes. Four saliency-based applications were developed after the validation of the saliency detection re- sults with a pseudo ground truth. The first and the second one concern respectively the optimal viewpoint selection and the adaptive compression of 3D non colored meshes. The third one shar- pens the details of a 3D colored mesh, and the fourth smooths adaptively its colors. The second aim is to propose a novel perceptual full reference metric for the quality assess- ment of 3D meshes. Given a 3D reference mesh (reputed devoid from any distorsions) and a 3D distorted mesh as inputs of this metric, the goal is to assess the perceived quality of the distorted mesh by providing a fidelity score which must be as close as possible to the humans scores. As vi- sual saliency is a pertinent information for our visual system, its use in the pipeline of the quality metric is natural. We use two properties of 3D meshes for evaluating their perceived quality : the visual saliency and the roughness. The multi-scale saliency map is used for the extraction of the structural informations of the 3D mesh and the roughness map for the account of the visual mas- king effect. We introduce 4 comparison functions between 2 corresponding local neighborhoods in order to estimate the structural differences between them. We combine these functions with a weighted Minkowski sum so as to obtain a final quality score. The third objectif of this thesis is to provide an approach for the difficult problem of the no reference qualityassessment of 3Dmeshes. On the contrary to full referencemetrics, this category is considered as the thorniest since the reference version of the 3D mesh is not used. Similarly to the full reference metric, we suppose that the visual quality of a 3D mesh is more affected when the salient areas of the mesh are affected and vice versa. We begin bysegmenting themesh into anumber of Superfacets which represent the local patches in this context. Then we affect to each vertex of a Superfacet its respective values of saliency and roughness. Afterward, we extract four local characteristics of each Superfacet(mean saliency, standard-deviation saliency, covariance saliency and mean roughness). Variations of these 4 cha- racteristics quantify effectively the distorsions that a mesh may undergo. Finally, we perform a learning step based onSVMs (Support VectorsMachines) using the constructed feature vector : to move from a vectorial representation to a final quality score, we use a regression SVM scheme.
Complete list of metadata
Contributor : Greyc Référent Connect in order to contact the contributor
Submitted on : Friday, December 16, 2016 - 4:01:44 PM
Last modification on : Saturday, June 25, 2022 - 9:51:19 AM
Long-term archiving on: : Monday, March 27, 2017 - 11:52:08 PM


  • HAL Id : tel-01418334, version 1


Anass Nouri. Cartes de saillance et évaluation de la qualité des maillages 3D . Vision par ordinateur et reconnaissance de formes [cs.CV]. Normandie Université, France, 2016. Français. ⟨tel-01418334⟩



Record views


Files downloads