Rotationally symmetric operations in the image domain may give rise to shape distortions. This article describes a way of reducing this effect for a general class of methods for deriving 3-D shape cues from 2-D image data, which are based on the estimation of locally linearized distortion of brightness patterns. By extending the linear scale-space concept into an affine scale-space representation and performing affine shape adaption of the smoothing kernels, the accuracy of surface orientation estimates derived from texture and disparity cues can be improved by typically one order of magnitude. The reason for this is that the image descriptors, on which the methods are based, will be relative invariant under a ne transformations, and the error will thus be confined to the higher-order terms in the locally linearized perspective mapping.

This article describes a method for reducing the shape distortions due to scale-space smoothing that arise in the computation of 3-D shape cues using operators (derivatives) de ned from scale-space representation. More precisely, we are concerned with a general class of methods for deriving 3-D shape cues from 2-D image data based on the estimation of locally linearized deformations of brightness patterns. This class

A unified framework for shape from texture and contour is proposed. It is based on the assumption that the surface markings are not systematically compressed, or formally, that they are weakly isotropic. The weak isotropy principle is based on analysis of the directional statistics of the projected surface markings. It builds on several previous theories, in particular by Witkin [25] and Kanatani [15]. It extends these theories in various ways, most notably to perspective projection. The theory also provides an exact solution to an estimation problem earlier solved approximately by Kanatani. The weak isotropy principle leads to a computationally efficient algorithm, WISP, for estimation of surface orientation. WISP uses simple image observables that are shown to be direct correlates of the surface orientation to compute an initial approximate estimate in a single step. In certain simple cases this first estimate is exact, and in experiments with natural images it is typically within 5...

Abstract not found

Existing techniques for obtaining shape from texture estimate tilt and slant via a single computational mechanism. These techniques do not take advantage of the fact that tilt is relatively easy to estimate, and slant can be estimated more easily once tilt is known. This paper introduces the adaptive bisector method, which allows tilt and slant to be calculated separately, and by different computational mechanisms. The method makes minimal assumptions regarding the isotropy of surface textures. Evidence from psychophysical studies suggests that human observers are able to provide accurate estimates of tilt[8], but are poor at estimating slant[4]. Whilst no psychophysical claims are made regarding the means by which tilt and slant are computed in this paper, it is claimed that a method which depends upon two different mechanisms provides a plausible functional model for how human observers obtain shape from texture. Results for real and synthetic perspective images of textured planar s...

This thesis introduces a novel representation for three-dimensional (3D) objects in terms of local affine-invariant descriptors of their appearance and the spatial relationships between the corresponding affine regions. Geometric constraints associated with different views of the same surface patches are combined with a normalized representation of their appearance to guide matching and reconstruction, allowing the acquisition of true 3D models from multiple unregistered images, as well as their recognition in photographs and image sequences. The proposed approach is applied to two domains: 1) Photographs – Models of rigid objects are constructed from photos and recognized in highly cluttered shots taken from arbitrary viewpoints. 2) Video – Dynamic scenes containing multiple moving objects observed by a moving camera are segmented into rigid components, and the 3D models constructed from these components are matched across different image sequences, with