Article Preview
Top1. Introduction
Indexing and retrieval techniques refer to a process of retrieving expected images from databases according to image queries or a number of features (e.g. color, texture and shape), which can be automatically extracted from the images using reasoning techniques. Some of those approaches use the outlines of objects described by a succession of straight line segments or curves (Belongie et Al., 2002; Zhang & Lu, 2004).
After many years of continuous development, indexing and retrieval technologies have become more and more mature and start play a key role in human life (e.g. a commercial package namely QBIC is available since 1995 (Flickner et Al., 1995)).
Various methods have been developed in order to represent and index shapes in an abstract and efficient way. The most interesting are part-based methods where a silhouette is decomposed into parts (Arandjelovic & Zisserman, 2010; Cronin, 2003), the aspect-graph methods that are viewer-centered representations of a three-dimensional object (Cyr & Kimia, 2004; Keysers et Al., 2007; Latecki et Al., 2005), methods that use the medial axis of silhouettes (Ma & Latecki, 2011; Mokhtarian, 1995) and appearance-based methods (Mokhtarian & Mackworth, 1992; Philbin et Al., 2007;Trinh & Kimia, 2011).
Indexing database of images facilitates the retrieval problem of any image. The indexing process has been considerably applied to various areas such as search engines, medical informatics, surveillance and digital rights management (Kato et Al., 1992; Hersh et Al., 2006; Hoad & Zobel, 2003).
The indexing and the recognition of objects using database of shapes is a current and difficult problem. Different methods based on different representations of shapes have been proposed. All of these methods proceed on descriptors comparison, and a similarity measure is then computed and used for the indexing process.
Among those methods, Geometry-based approaches that extract points from the image (usually edge or corner points) and reduce the problem to point set matching. Users are more interested in matching and retrieval by shape than by color and texture (Wang et Al., 2011; Yang et Al., 2008; Zaeri et Al, 2008). Segments are particularly interesting features (Lamdan & Wolfson, 1988; Matusiak, 1999; Lamiroy & Gros, 1996) because of their robustness to noise and their connectedness constraint that reduce the possibility of false matches. They also have the properties to vary slightly with a small change in the viewpoint, and to be invariant under similarity transform of the image (Gros, 1993).
Our aim is to develop an indexing and retrieval system. This system is based, first, on the extraction of elementary primitives of objects. The objects are represented by their silhouettes because our system is an appearance-based. This induces that the indexing process is restricted to 2D-2D matching, instead of directly interpreting 3D object information, and perform the object retrieval in the 2D indexes representation space.
In Sections 2 and 3, our modeling and indexing schemes are respectively presented and detailed. The proposed algorithms will be also given and discussed.
Our approach is validated in the fourth experimenting section in which application results on synthetic and real images will be presented and commented.