**Shape analysis** is the mainly automatic analysis of geometric shapes, for example using a computer to detect similarly shaped objects in a database or parts that fit together. For a computer to automatically analyze and process geometric shapes, the objects have to be represented in a digital form. Most commonly a boundary representation is used to describe the object with its boundary (usually the outer shell, see also 3D model). However, other volume based representations (e.g. constructive solid geometry) or point based representations (point clouds) can be used to represent shape.

Once the objects are given, either by modeling (computer-aided design), by scanning (3D scanner) or by extracting shape from 2D or 3D images, they have to be simplified before a comparison can be achieved. The simplified representation is often called a **shape descriptor** (or fingerprint, signature). These simplified representations try to carry most of the important information, while being easier to handle, to store and to compare than the shapes directly. A **complete shape descriptor** is a representation that can be used to completely reconstruct the original object (for example the medial axis transform).

### Shape descriptors

Shape descriptors can be classified by their invariance with respect to the transformations allowed in the associated shape definition. Many descriptors are invariant with respect to **congruency**, meaning that congruent shapes (shapes that could be translated, rotated and mirrored) will have the same descriptor (for example moment or spherical harmonic based descriptors or Procrustes analysis operating on point clouds).

Another class of shape descriptors (called **intrinsic** shape descriptors) is invariant with respect to isometry. These descriptors do not change with different isometric embeddings of the shape. **Their advantage is that they can be applied nicely to deformable objects** (e.g.__ a person in different body postures__) as these deformations do not involve much stretching but are in fact near-isometric. **Such descriptors are commonly based on geodesic distances measures along the surface of an object or on other isometry invariant characteristics such as the ****Laplace-Beltrami**** ****spectrum****.**

There are other shape descriptors, such as **graph-based** descriptors like the medial axis or the Reeb graph that capture geometric and/or topological information and simplify the shape representation but can not be as easily compared as descriptors that represent shape as a vector of numbers.

From this discussion it becomes clear, that different shape descriptors target different aspects of shape and can be used for a specific application. Therefore, depending on the application, it is necessary to analyze how well a descriptor captures the features of interest.

## No comments:

## Post a Comment