Shape-based invariant feature extraction for object recognition

Abstract

The emergence of new technologies enables generating large quantity of digital information including images; this leads to an increasing number of generated digital images. Therefore it appears a necessity for automatic systems for image retrieval. These systems consist of techniques used for query specification and retrieval of images from an image collection. The most frequent and the most common means for image retrieval is the indexing using textual keywords. But for some special application domains and face to the huge quantity of images, keywords are no more sufficient or unpractical. Moreover, images are rich in content; so in order to overcome these mentioned difficulties, some approaches are proposed based on visual features derived directly from the content of the image: these are the content-based image retrieval (CBIR) approaches. They allow users to search the desired image by specifying image queries: a query can be an example, a sketch or visual features (e.g., colour, texture and shape). Once the features have been defined and extracted, the retrieval becomes a task of measuring similarity between image features. An important property of these features is to be invariant under various deformations that the observed image could undergo. In this chapter, we will present a number of existing methods for CBIR applications. We will also describe some measures that are usually used for similarity measurement. At the end, and as an application example, we present a specific approach, that we are developing, to illustrate the topic by providing experimental results. © Springer-Verlag Berlin Heidelberg 2012.