Hessian-Based Affine Adaptation of Salient Local Image Features

Abstract

Affine covariant local image features are a powerful tool for many applications, including matching and calibrating wide baseline images. Local feature extractors that use a saliency map to locate features require adaptation processes in order to extract affine covariant features. The most effective extractors make use of the second moment matrix (SMM) to iteratively estimate the affine shape of local image regions. This paper shows that the Hessian matrix can be used to estimate local affine shape in a similar fashion to the SMM. The Hessian matrix requires significantly less computation effort than the SMM, allowing more efficient affine adaptation. Experimental results indicate that using the Hessian matrix in conjunction with a feature extractor that selects features in regions with high second order gradients delivers equivalent quality correspondences in less than 17% of the processing time, compared to the same extractor using the SMM.

Evaluation of two-view geometry methods with automatic ground-truth generation

“…In [21,22] pairs of cameras are used to collect data sets and their epipolar geometry is used to evaluate test systems. The ground truth epipolar geometry for each pair is computed automatically by accumulating correspondences over the entire high resolution image set acquired with that pair.…”

Introducing a globally consistent orbital‐based localization system

“…The eigenvalues of the Hessian matrix at each pixel can be computed, defining the curvature along each direction. The sticking out and well‐shaped configurations in one image are normally expressed by the extreme eigenvalues of the Hessian matrix . Thus, by analyzing the distribution of the eigenvalues and preserving the outliers, blobs can emerge.…”

“…The sticking out and wellshaped configurations in one image are normally expressed by the extreme eigenvalues of the Hessian matrix. 75 Thus, by analyzing the distribution of the eigenvalues and preserving the outliers, blobs can emerge. The standard deviation can be considered as the main parameter in the context of the scale space theory, 77 thus the higher the values…”

A matching algorithm based on global viewpoint difference rectification for framework imagery

“…With the improvement of pixel resolution of the remote sensing image, the requirement for precision is improving, which always needs to achieve at sub-pixel level or higher precision.Typical point matching algorithm's matching precision is not high [I] [2], and it cannot deal with greater perspective transform. Some affine invariant operator [3][4] [5] undertake normalized operation on the affine deformation in feature region, improving image matching performance in different perspectives, but bringing decline on feature distinguishing ability and positioning precision. Affine-SIFT algorithm [6] convert the angle into the change of "longitude" information to realize image correction in large viewpoint change, thus achieving high angle matching of o planar objects.…”