The Discrete Wavelet Transform (DWT) has a wide range of applications from signal processing to video and image compression. We show that this transform, by means the lifting scheme, can be performed in memory computation-efficient way on modern, programmable GPUs, which regarded as massively parallel coprocessors through NVidia's CUDA compute paradigm. three main hardware architectures for 2D DWT (row-column, line-based, block-based) are shown unsuitable implementation. Our CUDA-specific...

A novel, physically motivated deformable model for shape recovery and segmentation is presented. The model, referred to as the charged-particle (CPM), inspired by classical electrodynamics based on a simulation of charged particles moving in an electrostatic field. charges are attracted towards contours objects interest field, whose sources computed gradient-magnitude image. electric field plays same role potential forces snake while internal interactions modeled repulsive Coulomb forces. We...

A multiscale, morphological method for the purpose of shape-based object recognition is presented. connected operator similar to hat-transform defined, and two scale-space representations are built, using curvature function as underlying one-dimensional signal. Each peak valley extracted described by its maximum average heights extent represents an entry in top or bottom scale spaces. We demonstrate based on spaces three large data sets, a set diatom contours, silhouettes from MPEG-7...

We present a GPU-based framework for extracting surface and curve skeletons of 3D shapes represented as large polygonal meshes. use an efficient parallel search strategy to compute point-cloud their distance feature transforms (FTs) with user-defined precision. regularize by new geodesic tracing technique which is orders magnitude faster more accurate than comparable techniques. reconstruct the input from skeleton clouds using fast image-based method. also show how skeletal manifold...

Abstract The estimation of object orientation from RGB images is a core component in many modern computer vision pipelines. Traditional techniques mostly predict single per image, learning one-to-one mapping between and rotations. However, when objects exhibit rotational symmetries, they can appear identical multiple viewpoints. This induces ambiguity the problem, making map to rotations one-to-many fashion. In this paper, we explore several ways addressing problem. doing so, specifically...

Computing skeletons of 2D shapes, and medial surface curve 3D is a challenging task. In particular, there no unified framework that detects all types using single model, also produces multiscale representation which allows to progressively simplify, or regularize, skeleton types. this paper, we present such framework. We model detection regularization by conservative mass transport process from shape's boundary its skeleton, next finally the shape center. The resulting density field can be...

We study the evacuation dynamics of a crowd evacuating from complex geometry in presence fire as well slowly spreading smoke curtain. The is composed two kinds individuals: those who know layout building, and do not rely exclusively on potentially informed neighbors to identify path towards exit. aim capture effect knowledge environment has interaction between evacuees their residence time evolving smoke. Our approach genuinely multiscale - we employ two-scale model that able distinguish...

Simulating (elastically) deformable models that can collide with each other and the environment remains a challenging task. The resulting contact problems be elegantly approached using Lagrange multipliers to represent unknown magnitude of response forces. Typical methods construct solve Linear Complementarity Problem (LCP) obtain This requires inverse generalized mass matrix, which is generally hard for deformable-body problems. In this article, we tackle such by directly solving Mixed...

Abstract A general framework for automatic segmentation of diatom images is presented. This a critical first step in contour‐based methods identification diatoms by computerized image analysis. We review existing results, adapt popular to this difficult problem, and finally develop method that substantially improves results. based on the watershed from mathematical morphology, belongs class hybrid techniques. The novelty use connected operators computation selection markers, ingredient avoid...

A physically motivated method for surface reconstruction is proposed that can recover smooth surfaces from noisy and sparse data sets.No orientation information required.By a new technique based on regularized-membrane potentials the input sample points are aggregated, leading to improved noise tolerability outlier removal, without sacrificing much with respect detail (feature) recovery.After aggregating volumetric grid, novel, iterative algorithm used classify grid as exterior or interior...

Abstract We present a method for extracting complex manifolds with an arbitrary number of (self‐) intersections from unoriented point clouds containing large amounts noise. Manifolds are formed in three‐step process. First, small flat neighbourhoods all possible orientations created around points. Next, assembled into larger quasi‐flat patches, whose overlaps give the global connectivity structure cloud. Finally, curved extracted patch graph via multiple‐source flood fill. The can be...

Abstract Modern MRI measurements deliver volumetric and time‐varying blood‐flow data of unprecedented quality. Visual analysis these potentially leads to a better diagnosis risk assessment various cardiovascular diseases. Recent advances have improved the speed quality imaging considerably. Nevertheless, remains compromised by noise lack spatiotemporal resolution. Besides data, also numerical simulations are employed. These based on mathematical models specific features physical reality....

Abstract We propose a method for digital hair removal from dermoscopic images, based on threshold-set model. For every threshold, we adapt recent gap-detection algorithm to find hairs, and merge results in single mask image.We hairs this by combining morphological filters medial descriptors.We derive robust parameter values our over 300 skin images.We detail GPU implementation of show how it compares favorably with five existing methods, terms removing both long stubble various colors,...

We propose a novel, geometrically adaptive method for surface reconstruction from noisy and sparse point clouds, without orientation information. The employs fast convection algorithm to attract the evolving towards data points. force field in which is convected based on generalized Coulomb potentials evaluated an grid (i.e., octree) using fast, hierarchical algorithm. Formulating as problem velocity generated by offers number of advantages. Unlike methods compute distance set implicit...

Diffusion Tensor Imaging (DTI) allows to noninvasively measure the diffusion of water in fibrous tissue. By reconstructing fibers from DTI data using a fiber-tracking algorithm, we can deduce structure In this paper, outline an approach accelerating such algorithm Graphics Processing Unit (GPU). This which is based on calculation geodesics, has shown promising results for both synthetic and real data, but limited its applicability by high computational requirements. We present solution uses...