3D point of interest detection via spectral irregularity diffusion
- PDF / 3,449,525 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 56 Downloads / 246 Views
O R I G I N A L A RT I C L E
3D point of interest detection via spectral irregularity diffusion Ran Song · Yonghuai Liu · Ralph R. Martin · Paul L. Rosin
Published online: 24 April 2013 © Springer-Verlag Berlin Heidelberg 2013
Abstract This paper presents a method for detecting points of interest on 3D meshes. It comprises two major stages. In the first, we capture saliency in the spectral domain by detecting spectral irregularities of a mesh. Such saliency corresponds to the interesting portions of surface in the spatial domain. In the second stage, to transfer saliency information from the spectral domain to the spatial domain, we rely on spectral irregularity diffusion (SID) based on heat diffusion. SID captures not only the information about neighbourhoods of a given point in a multiscale manner, but also cues related to the global structure of a shape. It thus preserves information about both local and global saliency. We finally extract points of interest by looking for global and local maxima of the saliency map. We demonstrate the advantages of our proposed method using both visual and quantitative comparisons based on a publicly available benchmark. Keywords Mesh saliency · Points of interest · Laplacian · Eigendecomposition
1 Introduction Detection of points of interest on a 3D surface is a fundamental problem in computer vision and graphics, with applications to problems such as shape matching, shape-based retrieval, mesh simplification and viewpoint selection. Its R. Song () · Y. Liu Department of Computer Science, Aberystwyth University, Aberystwyth, UK e-mail: [email protected] R.R. Martin · P.L. Rosin School of Computer Science & Informatics, Cardiff University, Cardiff, UK
widespread applicability has led to a range of definitions in the literature. Usually, points of interest are considered to be points which are (i) locally distinctive, and (ii) likely to attract visual attention from a human. The former is merely related to the local geometry of the surface, and can be easily formulated once distinctiveness has been defined. However, the latter corresponds to the semantic relevance of points, and is hard to describe by a computational model even if this is often consistent to some degree with local geometric distinctiveness [3]. Generally, two different processes influence visual attention. One operates top-down and depends on the task at hand, or the cognitive and affective factors of the observer. The other operates bottom-up, and is driven by features within the input data. The relationship between topdown and bottom-up factors remains controversial. For example, can top-down interest counteract strong visual stimulus signals and shift attention away from bottom-up features? In this article, we focus on bottom-up point of interest detection. In our experiments, we will compare different bottom-up point of interest detectors based on humangenerated ground truth data, in the hope that we can fairly measure how much a detector relates to the real interests of human perception. 1.1 Related work E
Data Loading...
