Invariant Features from Interest Point Groups

Abstract

This paper approaches the problem of finding correspondences between images in which there are large changes in viewpoint, scale and illumination. Recent work has shown that scale-space 'interest points' may be found with good repeatability in spite of such changes. Furthermore, the high entropy of the surrounding image regions means that local descriptors are highly discriminative for matching. For descriptors at interest points to be robustly matched between images, they must be as far as possible invariant to the imaging process. In this work we introduce a family of features which use groups of interest points to form geometrically invariant descriptors of image regions. Feature descriptors are formed by resampling the image relative to canonical frames defined by the points. In addition to robust matching, a key advantage of this approach is that each match implies a hypothesis of the local 2D (projective) transformation. This allows us to immediately reject most of the false matches using a Hough transform. We reject remaining outliers using RANSAC and the epipolar constraint. Results show that dense feature matching can be achieved in a few seconds of computation on 1GHz Pentium III machines.

Shape recognition with edge-based features

“…A single descriptor has small discriminatory power but a set of spatially neighbouring features which preserve the geometric relations under arbitrary viewing conditions can unambiguously identify an object. This property was successfully used in numerous approaches [5,20,21,22]. Given an object to recognize, we can restrict the transformations that it can undergo.…”

“…Lowe [12] developed an efficient object recognition approach based on scale invariant features (SIFT). This approach was recently extended to sub-pixel/sub-scale feature localization [5]. In the context of scale invariant features Mikolajczyk and Schmid [14] developed a scale invariant interest point detector.…”

Viewpoint Induced Deformation Statistics and the Design of Viewpoint Invariant Features: Singularities and Occlusions

“…Viewpoint changes induce transformations more general than affine, but far less general than an arbitrary scrambling of feature positions. Our work concentrates on the case in between, following the steps of [25,4,[26][27][28]. 2 More specifically, [29,30] have proposed region descriptors for salient regions detected at or near occluding boundaries.…”

High-Resolution Feature Evaluation Benchmark

“…In [3], the evaluated detectors provide scale invariant properties. On the other hand, the localization accuracy of a scale invariant feature may be dependent on the detected scale [11], because its position error in a certain pyramid layer is mapped to the ground plane of the scale space pyramid. In high resolution data, more features are expected to be detected in higher scales of the image pyramid.…”