The Methodology Of Feature Extraction For Analyzing 3D Shapes And Visualizing Scientific Data

Recent innovation in 3D acquisition technology,such as computer tomography,magnetic resonance imaging,3D laser scanning,ultrasound,radar,and microscopy has enabled highly accurate digitization of complex 3D objects.Numerous scientific disciplines,such as neuroscience,mechanical engineering,and astrophysics,rely on the analysis and processing of such geometric data to understand intricate geometric structures and facilitate new scientific discoveries.Beside the 3D shape data sets,nowadays there is a growing need to be able to accurately and efficiently cope with scientific data,which is emerging tremendously fast throughout modern research and industry and is of particular interest as it represents a wide variety of physical phenomena,for instances heat diffusion,dynamic fluid flows,and motion of molecules.To be rounded up with the explosion information era,we see that what is challenging the comprehensive limitations of artificial intelligence is how to understand data,instead of how to acquire data.Methodologies of extracting features of data,including3 D shapes and volumetric data,have been proven to be critical to overcome those difficulties.By the term of feature extraction,we mean to construct feature transformations that map the source to a target,of which the key information,helping us understand data,is retained,nevertheless the useless is intentionally removed in the context.In other words,it bridges what we have to what we want.The achievements,elaborated in this work,fell into two camps: geometric features and topological features.Here,we introduce the first camp.In art drawing,tone is about giving definition,contrast,and depth,of which a broad range of different tonal values is the primary ingredient.However,in Pen-and-Ink illustration,a usual manipulation of shading was that the shades were locally suggested by combinations of individual strokes.In order to address this problem,we introduced global tone to capture shades over triangle meshes by computing the presented Gaussian visibility.Computing Gaussian visibility is efficient in both the time and space,yet it suffers from overlooking surface self-occlusions.Therefore,we proposed to alleviate the drawback by using the technique of Raycasting,and the improved results were applied to the application of mesh saliency.Likewise state-of-the-art work on mesh saliency,our above work was bottom-up and thus limited to extracting saliencies where the local contrast is high.Further,we investigated a top-down approach for detecting saliencies where the global contrast is high.Within the second camp,we proposes a graph-based local shape descriptor called Conformal Factor-weighted Medial Axis(CFMA)to capture regions of interest(ROI)specified by polygon boundaries.Differing from most existing local shape descriptors that often exploit the geometrical nature of shapes only,CFMA combines the topological feature of Medial Axis with the geometrical feature of surface curvature,leading it more advantageous over the prior work to agreeing with the human perception of resemblance.Finally,this work presented our key insights into the topological structure of 3D real,symmetric second-order tensor fields,of which the central goal is to put forward a novel computation model we call Deviatoric Eigenvalue Wheel(DEW).Overall,in terms of the format of data sources,we studied not only 3D surfaces but also scientific data.The findings are relevant to not only geometry but also topology,moreover the features being extracted are about not only global but also partial.The horizon of this work is broad,and we hope that it would be inspiring for the readers from different domains to catch up the progresses in extracting features of three-dimensional data in computer graphics.