Ben Kunsberg | Brown University (original) (raw)
Papers by Ben Kunsberg
2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, 2012
ABSTRACT Shape from shading is a classical inverse problem in computer vision. We introduce a nov... more ABSTRACT Shape from shading is a classical inverse problem in computer vision. We introduce a novel mathematical formulation for calculating local surface shape based on covariant derivatives, rather than the customary integral minimization or P.D.E approaches. Motivated by neurobiology, we introduce the shading flow field (the tangent map to the image isophotes) between the image and the surface levels. Just as in the perceptual organization of texture, we use the parallel transport of our shading flow field to move the isophote field at different points on the unknown surface to a single point, amassing restrictions on our surface curvatures. Under simplifying assumptions we solve exactly for the light source/surface pairs needed for a local image patch to have a given shading flow. The magnitude of the brightness gradient then restricts this family to a single light source and surface estimate pair, up to the concave/convex ambiguity and an additional elliptical/saddle ambiguity. Example calculations illustrate our approach.
SIAM Journal on Imaging Sciences, 2014
ABSTRACT Shape-from-shading (SFS) is a classical inverse problem in computer and human vision. Th... more ABSTRACT Shape-from-shading (SFS) is a classical inverse problem in computer and human vision. This shape reconstruction problem is inherently ill-posed. We show that the isophotes on smooth surfaces with Lambertian reflectance can be directly related to surface properties without consideration of the light source. Using techniques from modern differential geometry, we derive relationships between the curvature of the isophotes and the shape operator for the surface. Neurobiology motivates the geometric approach, and our calculations allow us to characterize the matching local family of surfaces that can result from any given shading patch. We illustrate the local ambiguity in several examples.
Journal of Vision, 2014
We use simple surfaces with a prominent ridge --the majority of the image structure is a ridge. O... more We use simple surfaces with a prominent ridge --the majority of the image structure is a ridge. Our lighting involves two separate sources, but we do not use the assumption anywhere.
Lecture Notes in Morphogenesis, 2014
Journal of Vision, 2014
Perceiving 3D shape involves processing and combining different cues, including texture, shading,... more Perceiving 3D shape involves processing and combining different cues, including texture, shading, and specular reflections. We have previously shown that orientation flows produced by the various cues provide fundamentally different information about shape, leading to complementary strengths and weaknesses (see Cholewiak & Fleming, VSS 2013). An important consequence of this is that a given shape may appear different, depending on whether it is shaded or textured, because the different cues reveal different shape features. Here we sought to predict specific regions of interest (ROIs) within shapes where the different cues lead to better or worse shape perception. Since the predictions were derived from the orientation flows, our analysis provides a key test of how and when the visual system uses orientation flows to estimate shape. We used a gauge figure experiment to evaluate shape perception. Cues included Lambertian shading, isotropic 3D texture, both shading and texture, and pseudo-shaded depth maps. Participant performance was compared to a number of image and scene-based perceptual performance predictors. Shape from texture ROI models included theories incorporating the surface's slant and tilt, second-order partial derivatives (i.e., change in tilt direction), and tangential and normal curvatures of isotropic texture orientation. Shape from shading ROI models included image based metrics (e.g., brightness gradient change), anisotropy of the second fundamental form, and surface derivatives. The results confirm that individually texture and shading are not diagnostic of object shape for all locations, but local performance correlates well with ROIs predicted by first and second-order properties of shape. The perceptual ROIs for texture and shading were well predicted via the mathematical models. In regions that were ROI for both cues, shading and texture performed complementary functions, suggesting that a common front-end based on orientation flows can predict both strengths and weaknesses of different cues at a local scale.
Shape from shading is a classical inverse problem in computer vision. This shape reconstruction p... more Shape from shading is a classical inverse problem in computer vision. This shape reconstruction problem is inherently ill-defined; it depends on the assumed light source direction. We introduce a novel mathematical formulation for calculating local surface shape based on covariant derivatives of the shading flow field, rather than the customary integral minimization or P.D.E approaches. On smooth surfaces, we show second derivatives of brightness are independent of the light sources and can be directly related to surface properties. We use these measurements to define the matching local family of surfaces that can result from any given shading patch, changing the emphasis to characterizing ambiguity in the problem. We give an example of how these local surface ambiguities collapse along certain image contours and how this can be used for the reconstruction problem.
Perceiving 3D shape from shading and texture requires combining different, but complimentary, inf... more Perceiving 3D shape from shading and texture requires combining different, but complimentary, information about shape features extracted from 2D images. Here, we sought to predict specific regions of interest (ROIs) within images – derived from orientation flows – where each cue leads to locally better or worse shape perception. This analysis assesses whether the visual system uses orientation flows to estimate shape. A gauge figure experiment was used to evaluate shape perception for 3D objects with Lambertian shading, isotropic texture, both shading and texture, and pseudo-shaded depth maps. Participant performance was compared to image and scene-based perceptual predictors. Shape from texture ROI models incorporated surface slant and tilt, second order partial derivatives, and tangential and normal curvatures of texture orientation. Shape from shading ROI models included image based metrics, anisotropy of the second fundamental form, and surface derivatives. Results confirmed that, individually, texture and shading are not diagnostic of object shape for all locations, but local performance correlates well with ROIs predicted by first and second order shape properties. In regions that were ROI for both cues, shading and texture performed complementary functions, suggesting a common front-end based on orientation flows locally predicts both strengths and weaknesses of cues.
Journal of Veterinary Medicine Series B, 2006
Disease clusters were retrospectively explored at national level using a geo-referenced dataset f... more Disease clusters were retrospectively explored at national level using a geo-referenced dataset from the 2001 Uruguayan Footand-Mouth Disease (FMD) epidemic. Disease location and time (first 11 epidemic weeks) were analysed across 250 counties (of which 160 were infected), without and with control for human mobility related factors (human population and road densities). The null hypothesis of random disease distribution over space and/or time was assessed with: (i) purely temporal; (ii) purely spatial; and (iii) space/time tests. At least within epidemic weeks 2 and 6, a principal disease cluster was observed in 33 contiguous counties (P < 0.01). Two secondary clusters, located at >100 km from each other, were also observed (P < 0.01). The purely spatial test that controlled for human population density identified two non-contiguous clusters (P < 0.01). Space and time analysis also revealed the same 33 counties as members of the principal cluster, of which 31 were also clustered when human population was controlled (P < 0.01). No clusters were reported by the spatial test when road density was assessed. The hypothesis that human mobility related factors autocorrelate with disease was empirically supported by two pieces of information: (i) removal of human population/road densities eliminated >93.9% of the counties included in the principal disease cluster; and (ii) statistically significant correlations (P < 0.05) were observed in the first three epidemic weeks between road density and the number of cases. Clusters where human population density was associated with 47% greater number of cases/sq. km than that of the principal cluster indicated possible roles as disease vectors (vector clusters). Selective control policy in vector clusters is recommended. Periodic (i.e. weekly) cluster and correlation analyses of both disease and other covariates may facilitate disease surveillance and help design space-specific control policy. U. S.
2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, 2012
ABSTRACT Shape from shading is a classical inverse problem in computer vision. We introduce a nov... more ABSTRACT Shape from shading is a classical inverse problem in computer vision. We introduce a novel mathematical formulation for calculating local surface shape based on covariant derivatives, rather than the customary integral minimization or P.D.E approaches. Motivated by neurobiology, we introduce the shading flow field (the tangent map to the image isophotes) between the image and the surface levels. Just as in the perceptual organization of texture, we use the parallel transport of our shading flow field to move the isophote field at different points on the unknown surface to a single point, amassing restrictions on our surface curvatures. Under simplifying assumptions we solve exactly for the light source/surface pairs needed for a local image patch to have a given shading flow. The magnitude of the brightness gradient then restricts this family to a single light source and surface estimate pair, up to the concave/convex ambiguity and an additional elliptical/saddle ambiguity. Example calculations illustrate our approach.
SIAM Journal on Imaging Sciences, 2014
ABSTRACT Shape-from-shading (SFS) is a classical inverse problem in computer and human vision. Th... more ABSTRACT Shape-from-shading (SFS) is a classical inverse problem in computer and human vision. This shape reconstruction problem is inherently ill-posed. We show that the isophotes on smooth surfaces with Lambertian reflectance can be directly related to surface properties without consideration of the light source. Using techniques from modern differential geometry, we derive relationships between the curvature of the isophotes and the shape operator for the surface. Neurobiology motivates the geometric approach, and our calculations allow us to characterize the matching local family of surfaces that can result from any given shading patch. We illustrate the local ambiguity in several examples.
Journal of Vision, 2014
We use simple surfaces with a prominent ridge --the majority of the image structure is a ridge. O... more We use simple surfaces with a prominent ridge --the majority of the image structure is a ridge. Our lighting involves two separate sources, but we do not use the assumption anywhere.
Lecture Notes in Morphogenesis, 2014
Journal of Vision, 2014
Perceiving 3D shape involves processing and combining different cues, including texture, shading,... more Perceiving 3D shape involves processing and combining different cues, including texture, shading, and specular reflections. We have previously shown that orientation flows produced by the various cues provide fundamentally different information about shape, leading to complementary strengths and weaknesses (see Cholewiak & Fleming, VSS 2013). An important consequence of this is that a given shape may appear different, depending on whether it is shaded or textured, because the different cues reveal different shape features. Here we sought to predict specific regions of interest (ROIs) within shapes where the different cues lead to better or worse shape perception. Since the predictions were derived from the orientation flows, our analysis provides a key test of how and when the visual system uses orientation flows to estimate shape. We used a gauge figure experiment to evaluate shape perception. Cues included Lambertian shading, isotropic 3D texture, both shading and texture, and pseudo-shaded depth maps. Participant performance was compared to a number of image and scene-based perceptual performance predictors. Shape from texture ROI models included theories incorporating the surface's slant and tilt, second-order partial derivatives (i.e., change in tilt direction), and tangential and normal curvatures of isotropic texture orientation. Shape from shading ROI models included image based metrics (e.g., brightness gradient change), anisotropy of the second fundamental form, and surface derivatives. The results confirm that individually texture and shading are not diagnostic of object shape for all locations, but local performance correlates well with ROIs predicted by first and second-order properties of shape. The perceptual ROIs for texture and shading were well predicted via the mathematical models. In regions that were ROI for both cues, shading and texture performed complementary functions, suggesting that a common front-end based on orientation flows can predict both strengths and weaknesses of different cues at a local scale.
Shape from shading is a classical inverse problem in computer vision. This shape reconstruction p... more Shape from shading is a classical inverse problem in computer vision. This shape reconstruction problem is inherently ill-defined; it depends on the assumed light source direction. We introduce a novel mathematical formulation for calculating local surface shape based on covariant derivatives of the shading flow field, rather than the customary integral minimization or P.D.E approaches. On smooth surfaces, we show second derivatives of brightness are independent of the light sources and can be directly related to surface properties. We use these measurements to define the matching local family of surfaces that can result from any given shading patch, changing the emphasis to characterizing ambiguity in the problem. We give an example of how these local surface ambiguities collapse along certain image contours and how this can be used for the reconstruction problem.
Perceiving 3D shape from shading and texture requires combining different, but complimentary, inf... more Perceiving 3D shape from shading and texture requires combining different, but complimentary, information about shape features extracted from 2D images. Here, we sought to predict specific regions of interest (ROIs) within images – derived from orientation flows – where each cue leads to locally better or worse shape perception. This analysis assesses whether the visual system uses orientation flows to estimate shape. A gauge figure experiment was used to evaluate shape perception for 3D objects with Lambertian shading, isotropic texture, both shading and texture, and pseudo-shaded depth maps. Participant performance was compared to image and scene-based perceptual predictors. Shape from texture ROI models incorporated surface slant and tilt, second order partial derivatives, and tangential and normal curvatures of texture orientation. Shape from shading ROI models included image based metrics, anisotropy of the second fundamental form, and surface derivatives. Results confirmed that, individually, texture and shading are not diagnostic of object shape for all locations, but local performance correlates well with ROIs predicted by first and second order shape properties. In regions that were ROI for both cues, shading and texture performed complementary functions, suggesting a common front-end based on orientation flows locally predicts both strengths and weaknesses of cues.
Journal of Veterinary Medicine Series B, 2006
Disease clusters were retrospectively explored at national level using a geo-referenced dataset f... more Disease clusters were retrospectively explored at national level using a geo-referenced dataset from the 2001 Uruguayan Footand-Mouth Disease (FMD) epidemic. Disease location and time (first 11 epidemic weeks) were analysed across 250 counties (of which 160 were infected), without and with control for human mobility related factors (human population and road densities). The null hypothesis of random disease distribution over space and/or time was assessed with: (i) purely temporal; (ii) purely spatial; and (iii) space/time tests. At least within epidemic weeks 2 and 6, a principal disease cluster was observed in 33 contiguous counties (P < 0.01). Two secondary clusters, located at >100 km from each other, were also observed (P < 0.01). The purely spatial test that controlled for human population density identified two non-contiguous clusters (P < 0.01). Space and time analysis also revealed the same 33 counties as members of the principal cluster, of which 31 were also clustered when human population was controlled (P < 0.01). No clusters were reported by the spatial test when road density was assessed. The hypothesis that human mobility related factors autocorrelate with disease was empirically supported by two pieces of information: (i) removal of human population/road densities eliminated >93.9% of the counties included in the principal disease cluster; and (ii) statistically significant correlations (P < 0.05) were observed in the first three epidemic weeks between road density and the number of cases. Clusters where human population density was associated with 47% greater number of cases/sq. km than that of the principal cluster indicated possible roles as disease vectors (vector clusters). Selective control policy in vector clusters is recommended. Periodic (i.e. weekly) cluster and correlation analyses of both disease and other covariates may facilitate disease surveillance and help design space-specific control policy. U. S.