Use of Texture for the Retrieval of Industrial Complexes in a Digital Image Database

III/age retrieval systems liavc been developed ill all attempt to increase the utilit» of digita! image databases. So III I' ill/age retrieval systems rely 011 charactcriring the texture of a digital ill/age. Current svstcms that use texture to search for man-made objects rely 011 pre-extraction and indexing. This studv attempt» to use a wavelet transform approach to retrieve industrial complexes front aerial photographs of Buffalo, Nell' York. A l laar wavelet transform is applied to database images and search ICO/IS. The wavelet transform produces a series ojsuuistical illdices that represent texture properties of the images. 7111' statistics of the search icon are compared to the statistics of all database images and the tell database regions with the greatest sunilaritv are retrieved. This studv explores the utility of image texture for retrieval of industrial complexes. Texture is evaluated tliroug]: retrieval perfonnance. A 50-907r retrieval accuracy is demonstrated. INTRODUCTION retrieving various land cover types and urban areas usi ng texture based wa velets. In general, searches in Image retrieval The production of digital spatial data has systems are conducted at two different levels of increased dramatically in the last decade. This abstraction semantic and feature vector (Castelli, volume of data production has created a need to 1998). In semantic level searches objects arc pre extracted and indexed (Castelli, 1998). ;\ semantic develop systems that can efficiently store and retrieve object can be defined as any part of an image to llJgital data to facilitate the use of this information. Image retrieval systems have been increasingly used which a semantic label can be assigned, such as to meet this need. building or road. A feature vector level search is one The purpose of image retrieval is to find that relies on statistical measures of the image images in a database that are VIsually similar to a produced by applying a wavelet transform. This sample image or query icon. Retrieval systems can approach uses sample images of known land­ use/land-cover types to retrieve similar images from operate on large databases and allow rctrievals to be an image database. Generally. areas not defined as performed on a variety of image properties such as texture and geometry (Castelli, 1998; Sheikholeslami objects, such as a forest, are searched for at the et al., 1997; Manjunath and Ma, 1996). The content feature vector level because they have uni form of the retrieval can be land use, structures or anything texture. contained within a set of digital images that has a Texture in digital images IS related to distinguishable texture or shape. properties such as shape, pattern, edges, tone and The potential utility of retrieval systems has pixel size (Avery and Berlin, 1992; Lillesand and been demonstrated III web based applications. Kiefer, 1994). In an aerial photograph, texture Digital libraries have been created which query reflects the incongruous pattern of the natural and spatially indexed data such as map and satellite built environment and usually requires a large areal images (Smith, 1996; Baxter and Anderson, 1996). extent to reveal a texture pattern. Therefore, texture­ Such systems have been successful in identifying and based wavelets applications have been successful in