Using Statistical Moment Invariants and Entropy in Image Retrieval

In many areas of commerce, government, academia, and hospitals, large collections of digital images are being created. Many of these collections are the product of digitizing existing collections of analogue photographs, diagrams, drawings, paintings, and prints. Usually, however, technologies related to archiving, retrieving, and editing images/video based on their content are still in their infancy, the only way of searching these collections was by keyword indexing, or simply by browsing. Digital image databases however, open the way to content-based searching. "Content-based" means that the search will analyze the actual contents of the image. The term content' in this context might refer to color, shape, texture, or any other information that can be derived from the image itself. Without the ability to examine image content, retrieval must rely on metadata such as captions or keywords, which may be laborious or expensive to produce.

What is desired is a similarity matching, independent of translation, rotation, and scale, between a given template (example) and images in the database. Consider the situation where a user wishes to retrieve all images containing cars, people, etc. in a large visual library. Being able to form queries in terms of sketches, structural descriptions, color, or texture, known as query-by-example (QBE), offers more flexibility over simple alphanumeric descriptions. This paper is organized as follows: section 2 provides the necessary background for CBIR. Section 3 defines the image segmentation technique using the Moments and Entropy. Section 4 explains the pro-posed model with the experiments conducted. Finally, section 5 concludes the paper and lists some further work.

Among the approaches used in developing early image database management systems (IDMS) are textual encoding [1], logical records [2], and relational databases [3]. The descriptions, employed to convey the content of the image, were mostly alphanumeric. Furthermore, these were obtained manually or by utilizing simple image processing operations designed for the application at hand. Later generations of IDMS have been designed in an object-oriented environment [4], where image interpretation routines form the backbone of the system. However, queries still remain limited to a set of predetermined features that can be handled by the system. The reader is referred to [5] for a survey of IDMS.

Most recent systems reported in the literature for searching, organizing, and retrieving images based on their content include IBMs Query-by-Image-Content (QBIC) [6], MITs photo-book [7], the Trademark and Art Museum applications from ETL [8], Xenomania from the University of Michigan [9], and Multimedia/VOD test bed applications from the Columbia University [10]. IBMs QBIC is a system that translates visual information into numerical descriptors, which are then stored in a database. It can index and retrieve images based on average color, histogram color, texture, shape, and sketches. MITs photobook describes three content-based tools, utilizing principal component analysis, finite element modes, and the Wold transform to match appearances, shapes, and textures, respectively, from a database to a prototype at run time. Xenomania is a system for face image retrieval, which is based on QBE. Its embedded routines allow for segmentation and evaluation of objects based on domain knowledge, yielding feature values that can be utilized for similarity measures and image retrieval. The database management system of the Columbia University proposes integrated feature maps based on texture, color, and shape information for image indexing and query in transform domain. Similarity-based searching in medical image databases has been addressed in [11]. A variety of shape representation and matching techniques are currently available, which are invariant to size, position, and/or orientation. They may be grouped as: (1) methods based on local features such as points, angles, line segments, curvature, etc. [12]; (2) template matching methods [13]; (3) transform coefficient based methods, including Fourier descriptors [14] or generalized Hough transform [15]; (4) methods using 3 modal and finite element analysis [16]; (5) methods based on geometric features, such as local and differential invariants [17]; and (6) methods using B-Splines or snakes for contour representation [18]. Comprehensive surveys of these methods can be found in [19].

A. Image Segmentation The shape representation method described here assumes that the object has been fully segmented from the original image, such that all pixels representing the objects shape have been identified as distinct from those pertaining to the rest of the image. In this paper, a local diffusive segmentation method [20] is used. There exist a wide variety of ways to achieve segmentation; however, it is not the subject of this paper. All contiguous pixels, which share a given point-based characte

…(Full text truncated)…

This content is AI-processed based on ArXiv data.