Rajiv Mehrotra’s research while affiliated with University of Kentucky and other places

With recent advances in computer technologies, numerous new application areas requiring management of non-alphanumeric data such as images, videos, graphs, and audios have evolved. Examples of such applications include weather information management, medical information management, environmental pollution information systems, space exploration, manufacturing information management, genome research, training and educational systems, entertainment applications, and defense applications.

An abstract is not available.

The data acquired by the ever-increasing utilization of a variety of sophisticated sensors in the automation of industrial and manufacturing tasks have attracted attention of the scientific community to the problems of storage, analysis and management of nontraditional data. Sensors can be and are being used for acquiring data necessary for automatic and intelligent control, analysis and decision-making for a number of industrial/manufacturing operations. Some industrial tasks that need sensor data include monitoring the progress of operations, inspection and recognition of tools/products, eye-hand coordination in assembly robots, automatic handling and sorting of material, monitoring wear and tear of tools during operation, automatic setting of machines (loading and unloading tools), diagnostics and maintenance of equipments, and autonomous vehicles, to name a few. In a large number of sensor data-based industrial operations, the collected sensor data are required to be stored for further analysis. For instance, sensor data related to tools’ wearing and tearing and the product inspection data can be used to investigate the relationships among the tool defects and the product defects. Similarly, the sensor data gathered by monitoring a metal cutting operation can be used to study the chip curling and breaking behavior, which could lead to better metal cutting techniques. In some cases, the acquired sensor data are needed to retrieve stored data for determining the task or a sequence of tasks required to be performed. For example, in a completely flexible and automatic assembly system, sensor data are used to identify and determine the poses of one or more specific parts on the conveyor belt.

Contemporary database management systems are devised to give users a seamless and transparent view into the data landscape being managed. Such programs give users the illusion that their view of the data corresponds to the way that it is actually internally represented, as if they were the only users of the software. Image database management system was conceived as a way of managing images for image algorithm development test beds. Images were retrieved based on information in header files, which contained only textual information. The architecture of a standard database management system is usually divided into three different levels, corresponding to the ANSI/SPARC standard. These levels are the physical database level, the conceptual database level, and the external database level. The implementation-independent framework that is employed to describe a database at the logical and external level is called a data model. These models represent the subject database in terms of entities, entity types, attributes of entities, operations on entities and entity types, and relationships among entities and entity types. An image data model must represent the following types of information: the model base, the model-base instantiation, the instantiation-object connection, and the object information repository. The chapter discusses some examples of the existing image database management systems classifying them as first, second, and third generation database systems. It also focuses on the similarity retrieval in image database systems.

We are currently developing unified query processing strategies for image databases. To perform this task, model-based representations of images by content are being used, as well as a hierarchical generalization of a relatively new object-recognition technique called data-driven indexed hypotheses. As the name implies, it is index-based, from which its efficiency derives. Earlier approaches to data-driven model-based object recognition techniques were not capable of handling complex image data containing overlapping, partially visible, and touching objects due to the limitations of the features used for building models. Recently, a few data-driven techniques capable of handling complex image data have been proposed. In these techniques, as in traditional databases, iconic index structures are employed to store the image and shape representation in such a way that searching for a given shape or image feature can be conducted efficiently. Some of these techniques handle the insertion and deletion of shapes and/or image representations very efficiently and with very little influence on the overall system performance. However, the main disadvantage of all previous data-driven implementations is that they are main memory based. In the present paper, we describe a secondary memory implementation of data-driven indexed hypotheses along with some performance studies we have conducted.

This analysis is a panel discussion. There are many problems in the field of image database management. The object-oriented paradigm has been and continues to be a great impetus to this work. The semantics of images is essentially what they contain, and unless there is an effective method to identify their contents and index them on that basis, the database will degenerate to a collection of patterns with no semantics. This is the most challenging issue facing multimedia information systems in general, and image databases in particular. Work on query by image content has barely begun to scratch the surface. A few key query primitives will become well-understood and widely supported. To allow users to browse and search through information domains using sophisticated querying techniques that include imprecise queries, user-directed query processing, and queries that use similarity measures in order to retrieve data, new data modeling approaches are required. Key problems that arise in providing query by image content are considered

Corner detection is a low-level feature detection operator that is of great use in image processing applications, for example, optical flow and structure from motion by image correspondence. The detection of corners is a computationally intensive operation. Past implementations of corner detection techniques have been restricted to software. In this paper we propose an efficient very large-scale integration (VLSI) architecture for detection of corners in images. The corner detection technique is based on the half-edge concept and the first directional derivative of Gaussian. Apart from the location of the corner points, the algorithm also computes the corner orientation and the corner angle and outputs the edge map of the image. The symmetrical properties of the masks are utilized to reduce the number of convolutions effectively, from eight to two. Therefore, the number of multiplications required per pixel is reduced from 1800 to 392. Thus, the proposed architecture yields a speed-up factor of 4.6 over conventional convolution architectures. The architecture uses the principles of pipelining and parallelism and can be implemented in VLSI.

Corners are special features in images, and are of great use in computing the optical flow and structure from motion. Conventionally, corners have been defined as the junction point of two straight line edges. Most existing edge detectors perform poorly at corners, mainly because they assume an edge to be an entity having infinite extent, which is violated at the corners. Since, most of the corner detectors are based on existing edge detectors, the performance of such corner detectors is not satisfactory. A corner point can be viewed as the intersection of two half-edges, oriented in 2 different directions, which are not 180° apart. This statement defines both a half-edge and a corner in terms of one another. This definition is the essence of the corner detection strategies presented in this paper. The corner detection algorithms rely on detecting half-edges. A half-edge detector uses information from a single orientation rather than opposing directions. We propose two algorithms for edge detection and corner detection, one is based on the First Directional Derivative of Gaussian and the other is based on the Second Directional Derivative of Gaussian. In addition to the location of the corner points, our algorithms also determine the corner angle and the corner orientation. The efficacy of the detectors has been demonstrated by experimental results for laboratory scenes.