Prime Shapes in Natural Images

Abstract

This paper provides evidence that about half of all the regions in segmented images can be classified as one a few simple shapes. Using three segmentation algorithms, three different image databases, and two shape descriptors, we empirically show that shapes such as triangles, squares, and circles are observed, up to an affine transform and at a much higher rate than random shapes. This result has potential value in applications such as scene understanding, visual object classification, and matching because qualitative shapes can be used as features. We show an application in scene categorisation based on what might be called 'bag of shapes' .

Cross-depiction problem: Recognition and synthesis of photographs and artwork

“…Shrivista et al [33] show that an exemplar SVM trained on a huge database is capable of classification of both photographs and artwork. A less computationally intensive approach has been proposed [34] using a hierarchical graph model to obtain a coarse-to-fine arrangement of parts with nodes labelled by qualitative shape [35]. Wu et al [36] address the cross-depiction problem using a deformable model; they use a fully connected graph with learned weights on nodes (the importance of nodes to discriminative classification), on edges (by analogy, the stiffness of a spring connecting parts), and multiple node labels (to account to different depictions); a method tested on 50 categories.…”

Modelling Visual Objects Invariant to Depictive Style

“…However, rather than using complicated shapes for regions (as others do), or just using (a hierarchy of) Gaussian blobs [18], we use a collection of simple shapes (e.g. circle, square, triangle) [25]. The idea is that abstracting region shape into one of a few classes brings greater robustness to non-salient variations.…”

“…The shape-class prior, p(S) is taken to be the relative number of shapes classified as shape S. All parameters used are provided by the shape classifier after training on about 40000 regions [25]. Figure 3 illustrates the shape classes we use, and the shape classes used to label nodes at each level of a hierarchy.…”

“…The first five of these shapes have been shown to explain around 80% of regions in photographs up to an affine transform; random is used when a region cannot be classified. See [25] for details on shape classification. "Polygon" captures pentagons, hexagons, etc.…”

Depiction Inviariant Object Matching

“…However, the particular set of shapes used were presumed to be useful by the algorithm designers, and finding the optimal qualitative shape for any given object part is expensive and not straight forward. Recent work has solved both of these problems [19]. Impressive results have been obtained for matching similar images of different depictions based on learning [14].…”