Abstract

In this work an overview of the most important methods for interest-point-detection is provided. Common concepts are explained in detail. Furthermore, a framework is implemented which integrates common interest-point-detectors and facilitates a visual comparison of the results of the different techniques. A set of images composed of simple structures is allocated for the comparison.
Many applications in the field of computer vision which rely on computer-aided processing of image data, for example tracking, object recognition and 3D-reconstruction depend on the detection of interesting structures in images prior to further processing steps to allow for robust image matching. These structures usually correspond to so-called blobs or corners and are referred to as local features. The position of local features in images is determined by so-called interest-points or interest-regions. Techniques to identify local features in images are called interest-point-detectors. Methods based on interest-point-detection have proven to be especially well-suited for robust image matching and are therefore most commonly used in current computer vision systems. In contrast to other approaches, methods based on interest-points use local image information for the detection of the aforementioned features. Thus, they produce good results even when objects in images are partially occluded. Furthermore, local features determined by interest-point-detectors are robust to various geometric and photometric image transformations and yield a compact representation of image content.
Research on methods for the detection of robust interest-points has been done since the late seventies. Accordingly there is a great number of different interest-point-detectors nowadays. In order to be able to choose the appropriate interest-point-detector for a specific task it is vital to familiarize oneself with the basic concepts and methods of interest-point-detection.

Reference

Eisele, P. (2011). Evaluierung und Visualisierung von Interest-Point-Detektoren [Diploma Thesis, Technische Universität Wien]. reposiTUm. https://resolver.obvsg.at/urn:nbn:at:at-ubtuw:1-51393