<title>Retrieving images by 2D shape: a comparison of computation methods with human perceptual judgments</title>

Abstract

In content based image retrieval, systems allow users to ask for objects similar in shape to a query object. However, there is no clear understanding of how computational shape similarity corresponds to human shape similarity. In this paper several shape similarity measures were evaluated on planar, connected, non-occluded binary shapes. Shape similarity using algebraic moments, spline curve distances, cumulative turning angle, sign of curvature and Hausdorff-distance were compared to human similarity judgments on twenty test shapes against a large image database.

Image Retrieval: Current Techniques, Promising Directions, and Open Issues

“…QBIC [99,48,46,44,129,74,41], standing for query by image content, is the first commercial content-based image retrieval system. Its system framework and techniques have profound effects on later image retrieval systems.…”

“…combinations of coarseness, contrast, and directionality [44]. Its shape feature consists of shape area, circularity, eccentricity, major axis orientation, and a set of algebraic moment invariants [129,46]. QBIC is one of the few systems which takes into account the high dimensional feature indexing.…”

Rotation invariant distance measures for trajectories

“…Moreover, while this method can handle global scaling, there is no support for local scaling. Compared to other methods based on algebraic moments, curvature [11] and Hausdorff distance [3], the turning angle has been shown to be the most robust for the retrieval of two dimensional images and shapes [12].…”

Silhouette lookup for monocular 3D pose tracking

“…The turning angle metric is sensitive to the length and orientation of extended limbs, and has been shown to correlate well with human notions of shape similarity (12). In brief, the turning angle metric measures the integral of the difference between two normalized functions, where each function is derived from a silhouette by taking the tangent trace made during one complete circuit around the silhouette's border (see Figure 3).…”