A rotation- and flip-invariant algorithm for representing spatial continuity information of geographic images in content-based image retrieval

Abstract

This research proposes a rotation- and flip-invariant algorithm for representing spatial continuity information in high-resolution geographic images for content based image retrieval (CBIR). Starting with variogram concept, the new visual property representation, in the form of a numeric index vector, consists of a set of semi-variances at selected lags and directions, based on three well-justified principles: (1) capture the basic shape of sample variogram, (2) represent the spatial continuity anisotropy, and (3) make the representation rotation- and flip-invariant. The algorithm goes through two tests. The first test confirms that it can indeed align the image representations based on spatial continuity information of objects within images by re-ordering the semi-variances accordingly. In the second test, the algorithm is applied to retrieve seven types of typical geographic entities from an Erie County orthophoto database. The retrieval results demonstrate the effectiveness of the new algorithm in CBIR, as assessed by retrieval precision.

Introduction

Content based image retrieval (CBIR) has gained increasing attention from GIS scientists (Manjunath and Ma, 1996; Bruns and Egenhofer, 1996; Sheikholeslami et al., 1999; Agouris et al., 1999; Stefanidis et al., 2002). Originating from the computer vision and database community (Flickner et al., 1995; Pentland et al., 1996; Smith and Chang, 1996), CBIR attempts objectively and efficiently to retrieve targeted images or image regions from a large-volume image database based on the content similarity between a query icon (or sample image) and database images. What distinguishes it from conventional image retrieval is that in CBIR, image content is represented using numeric measurements of visual properties such as color, texture and shape (Flickner et al., 1995; Pass and Zabih, 1996; Rui et al., 1999).

While the interest in CBIR for analysis of geographic-image databases are rising, more attention needs to be on the uniqueness of geographic images and appropriate visual content representations. Recent CBIR research in geographic applications mostly focused on developing general CBIR approaches and applying them in retrieving texture-rich geographic entity types, such as forest, farmland, grasses, and others (Manjunath and Ma, 1996; Sheikholeslami et al., 1999). Man-made objects, such as factories, shopping centers and roads, in contrast, are much less studied, despite the fact that retrieving these man-made objects has important and broad applications, especially for urban areas. These man-made objects are discrete, but embedded in a complex background, and may not have rich texture patterns: thus retrieving these man-made objects presents a challenge conceptually as well as methodologically. Some relevant researches along this direction are reported recently (Bruns and Egenhofer, 1996; Agouris et al., 1999; Stefanidis et al., 2002), and they are based on object layers and assume that appropriate delineation of such objects has been completed either from geographic images or other sources. However, for images with complex scene structures like geographic images, automatic object delineation is still not an easy task, if not impossible. This paper presents an algorithm effectively to represent geographic-image content for CBIR, and its representation is directly derived from raw images. The algorithm is based on a well-recognized property of geographic space—spatial continuity.

In the following sections, a short discussion is first devoted on the process of general CBIR and its relationships with image classification. We then briefly review the concept of spatial continuity and its relevance in geographic-image study. Next, the new spatial continuity-based algorithm for CBIR is presented and experimental results are reported. Finally, we draw brief conclusions and make some discussions