We introduce two criteria to regularize the optimization involved in learning a classifier domain where no annotated data are available, leveraging different domain, problem known as unsupervised adaptation. focus on task of semantic segmentation, synthetic aplenty, but annotating real is laborious. The first criterion, inspired by visual psychophysics, that map between image domains be phase-preserving. This restricts set possible learned maps, while enabling enough flexibility transfer...

We define a novel metric on the space of closed planar curves which decomposes into three intuitive components. According to this metric, centroid translations, scale changes, and deformations are orthogonal, is also invariant with respect reparameterizations curve. While earlier related Sobolev metrics for exhibit some general similarities proposed in work, they lacked important three-way orthogonal decomposition, has particular relevance tracking computer vision. Another positive property...

Recently proposed Sobolev active contours introduced a new paradigm for minimizing energies defined on curves by changing the traditional cost of perturbing curve and thereby redefining their gradients. evolve more globally are less attracted to certain intermediate local minima than contours, it is based well-structured Riemannian metric. In this paper, we analyze using scale-space analysis in order understand evolution across different scales. This shows an extremely important useful...

We define a manifold $M$ where objects $c\in M$ are curves, which we parameterize as $c:S^1\to R^n$ ($n\ge 2$, $S^1$ is the circle). Given curve $c$, tangent space $T_cM$ of at $c$ including in it all deformations $h:S^1\to $c$. In this paper study geometries on provided by Sobolev--type metrics $H^j$. $H^j$ type for cases $j=1,2$; prove estimates, and characterize completion smooth curves. As bonus, that Fr\'echet distance curves (see arXiv:math.DG/0312384) coincides with induced ``Finsler...

Ultrafiltration asymmetric porous membranes were imaged by two microscopy methods, which allow 3D reconstruction: focused ion beam and serial block face scanning electron microscopy. A new algorithm was proposed to evaluate porosity average pore size in different layers orthogonal parallel the membrane surface. The reconstruction enabled additionally visualization of interconnectivity parts membrane. method demonstrated for a copolymer can be extended other with application ultrafiltration,...

We consider regions of images that exhibit smooth statistics, and pose the question characterizing "essence" these matters for recognition. Ideally, this would be a statistic (a function image) does not depend on viewpoint illumination, yet is sufficient task. In manuscript, we show such statistics exist. That is, one can compute deterministic functions image contain all "information" present in original image, except effects illumination. also are supported "thin" (zero-measure) subset...

This work provides a model for tubular structures, and devises an algorithm to automatically extract anatomical structures from medical imagery. Our fits many in imagery, particular, various fiber bundles the brain (imaged through diffusion-weighted magnetic resonance (DW-MRI)) such as cingulum bundle, blood vessel trees computed tomography angiograms (CTAs). Extraction of bundle is interest because possible ties schizophrenia, extracting vessels helpful diagnosis cardiovascular diseases....

We propose a method to detect disocclusion in video sequences of three-dimensional scenes and partition the disoccluded regions into objects, defined by coherent deformation corresponding surfaces scene. Our infers fields that are piecewise smooth construction without need for an explicit regularizer associated choice weight. It then partitions region groups its components with objects leveraging on complementarity motion appearance cues: Where changes within object, can usually be reliably...

Recently proposed Sobolev active contours introduced a new paradigm for minimizing energies defined on curves by changing the traditional cost of perturbing curve and thereby redefining their gradients. evolve more globally are less attracted to certain intermediate local minima than contours. In this paper we analyze in Fourier domain order understand evolution across different scales. This analysis shows an important useful behavior contours, namely, that they move successively from coarse...

We consider several variants of the active contour model without edges, extended here to case noisy and blurry images, in a multiphase multilayer level set approach. Thus, models jointly perform denoising, deblurring segmentation variational formulation. To minimize practice proposed functionals, one most standard ways is use gradient descent processes, time dependent Usually, L² functional computed discretized practice, based on inner product. However, this computation often...

We present a method to track the shape of an object from video. The uses joint and appearance model object, which is propagated match radiance in subsequent frames, determining shape. Self-occlusions dis-occlusions camera motion pose difficulties models tracking. They are unable adapt new information, leading inaccurate detection. In this work, we self-occlusions tracking framework. warp propagate coupled, thus formulate optimization problem. derive coarse-to-fine method, advantageous...

Active contour and active polygon models have been used widely for image segmentation. In some applications, the topology of object(s) to be detected from an is known a priori, despite unknown complex geometry, it important that or maintain desired topology. this work, we construct novel geometric flow can added based evolutions contours polygons so initial preserved. Indeed, proposed ensures more than just correct topology; is, in sense, kept far away change. Smoothness properties similar...

We propose a dynamical model-based approach for tracking the shape and deformation of highly deforming objects from time-varying imagery. Previous works have assumed that object is smooth, which realistic problem, but most restricted to belong finite-dimensional group, such as affine motions, or finitely-parameterized models. This, however, limits accuracy scheme. exploit smoothness assumption implicit in previous work, we lift restriction motions/deformations. To do so, derive analytical...

We present a method to track the precise shape of dynamic object in video. Joint and appearance models, which template is propagated match radiance next frame, are advantageous over methods employing global image statistics cases complex cluttered background. In 3D motion relative viewpoint change, self-occlusions dis-occlusions prominent, current joint models unable accurately adapt new information, leading inaccurate detection. this work, we model tracking framework. Experiments on video...

We consider the problem of filling in missing spatiotemporal regions a video. provide novel flow-based solution by introducing generative model images relation to scene (without regions) and mappings from images. use jointly infer template, 2D representation scene, mappings. This ensures consistency frame-to-frame flows generated underlying reducing geometric distortions flow based inpainting. The template is mapped video new (L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML"...

We present a region-based active contour detection algorithm for objects that exhibit relatively homogeneous photometric characteristics (e.g. smooth color or gray levels), embedded in complex background clutter. Current methods either frame this problem Bayesian classification terms, where precious modeling resources are expended representing the away from decision boundaries, use heuristics to limit search local regions around object of interest. propose an adaptive lookout region, whose...

We propose new dense descriptors for texture segmentation. Given a region of arbitrary shape in an image, these are formed from shape-dependent scale spaces oriented gradients. These defined by Poisson-like partial differential equations. A key property our is that they do not aggregate image data across the boundary region, contrast to existing based on aggregation As example, we show how descriptor can be incorporated Mumford-Shah energy test method several challenging datasets...

We formulate an energy for segmentation that is designed to have preference segmenting the coarse over fine structure of image, without smoothing across boundaries regions. The formulated by integrating a continuum scales from scale space computed heat equation within show can be optimized computing scales, but instead single scale. This makes method computationally efficient in comparison energies using discrete set scales. apply our texture and motion segmentation. Experiments on benchmark...

We consider the problem of optimization cost functionals on infinite dimensional manifold diffeomorphisms. present a new class methods, valid for any setup space diffeomorphisms by generalizing Nesterov accelerated to While our framework is general manifolds, we specifically treat case diffeomorphisms, motivated optical flow problems in computer vision. This accomplished building recent variational approach methods Wibisono, Wilson, and Jordan [Proc. Natl. Acad. Sci. USA, 113 (2016), pp....

We present a general framework and method for detection of an object in video based on apparent motion. The moves relative to background motion at some unknown time the video, goal is detect segment as soon it online manner. Due unreliability between frames, more than two frames are needed reliably object. Our designed object(s) with minimum delay, i.e., after moves, constraining false alarms. Experiments new extensive dataset moving show that our achieves less delay all alarm constraints...