Sanjiv K. Bhatia

Conceptual clustering in information retrieval

Clustering is used in information retrieval systems to enhance the efficiency and effectiveness of the retrieval process. Clustering is achieved by partitioning the documents in a collection into classes such that documents that are associated with each other are assigned to the same cluster. This association is generally determined by examining the index term representation of documents or by capturing user feedback on queries on the system. In cluster-oriented systems, the retrieval process can be enhanced by employing characterization of clusters. In this paper, we present the techniques to develop clusters and cluster characterizations by employing user viewpoint. The user viewpoint is elicited through a structured interview based on a knowledge acquisition technique, namely personal construct theory. It is demonstrated that the application of personal construct theory results in a cluster representation that can be used during query as well as to assign new documents to the appropriate clusters.

Image database indexing using JPEG coefficients

Abstract Image database indexing is used for efficient retrieval of images in response to a query expressed as an example image. The query image is processed to extract information that is matched against the index to provide pointers to similar images. We present a technique that facilitates content similarity-based retrieval of jpeg -compressed images without first having to uncompress them. The technique is based on an index developed from a subset of jpeg coefficients and a similarity measure to determine the difference between the query image and the images in the database. This method offers substantial efficiency as images are processed in compressed format, information that was derived during the original compression of the images is reused, and extensive early pruning is possible. Initial experiments with the index have provided encouraging results. The system outputs a set of ranked images in the database with respect to the query using the similarity measure, and can be limited to output a specified number of matched images by changing the threshold match.

Selection of search terms based on user profile

A number of researchers have addressed the problem of selecting additional search terms for a query using term dependency information. Some other techniques rely on the construction of user profiles and the identification of individual users with a class of users having known backgrounds. In other words, these latter techniques for the construction of user profiles do not consider user preferences at the level of individual users. In this paper, we present a new technique for the construction of a user profile in the form of a concept (construct) dependence tree. This technique makes use of personal construct theory that involves active user participation. The user profile captures the user preferences and may be used to select additional search terms. Our initial results indicate that the dependence tree generated by this technique can improve retrieval effectiveness.

Perfect hash table algorithm for image databases using negative associated values

Abstract A 2D string data structure allows for efficient spatial reasoning on an image database for query and retrieval. A 2D string can be converted to a set of triples leading to an elegant O (1) solution for image retrieval with simple queries using a perfect hash table. For complex queries, the retrieval complexity is linear in this approach and depends on the number of possible pairings of picture objects in the query. The perfect hash table computation for this problem is mapped directly to a permutation problem. In an earlier paper [S. K. Bhatia and C. L. Sabharwal, Pattern Recognition 27 , 365–376 (1974)], we presented a set of heuristics that result in a fast computation of associated values, for picture objects, used in the calculation of hash addresses. In this paper, we present an additional heuristic leading to a 90% reduction in search space over our earlier algorithm. The new heuristic promises to generate a minimal perfect hash function for each experimental data set, which was not possible with the earlier algorithms. Mathematical analysis of complexity of the algorithms is presented and is supported by experimental results.

Searching satellite imagery with integrated measures

Due to the advances in imaging and storage technologies, the number and size of images continue to grow at a rapid pace. This problem is particularly acute in the case of remotely sensed imagery. The continuous stream of sensory data from satellites poses major challenges in storage and retrieval of the satellite imagery. In the mean time, the ubiquity of Internet has resulted into an ever-growing population of users searching for various forms of information. In this paper, we describe the search engine SIMR-Satellite Image Matching and Retrieval system. SIMR provides an efficient means to match remotely sensed imagery. It computes spectral and spatial attributes of the images using a hierarchical representation. A unique aspect of our approach is the coupling of second-level spatial autocorrelation with quad tree structure. The efficiency of the web-based SIMR has been evaluated using a database of images with known characteristics: cities, towns, airports, lakes, and mountains. Results show that the integrated signature can be an effective basis for accurately searching databases of satellite based imagery.

Human Face Perception in Degraded Images

Abstract This paper reports human performance data from a series of psychophysical experiments investigating the limits of stimulus parameters relevant to distinguishing a human face in a mug shot. In these experiments, we use a two-alternative forced-choice paradigm for response elicitation. We develop a benchmark that can be used to determine the performance of a machine vision system for human face detection at different levels of image degradation. The benchmark is developed in terms of the number of pixel blocks and the number of gray scales used in the images. The paper presents a model of representation that can be useful for recognition of faces in a database, and may be used to define the minimum image quality required for retrieval of facial records at different confidence levels. Our results show that low-frequency information in face images is useful since it is most resilient to degradation in the image quality. The model is particularly relevant to the retrieval of facial images in large image databases.

A survey of image processing techniques and statistics for ballistic specimens in forensic science

This paper provides a review of recent investigations on the image processing techniques used to match spent bullets and cartridge cases. It is also, to a lesser extent, a review of the statistical methods that are used to judge the uniqueness of fired bullets and spent cartridge cases. We review 2D and 3D imaging techniques as well as many of the algorithms used to match these images. We also provide a discussion of the strengths and weaknesses of these methods for both image matching and statistical uniqueness. The goal of this paper is to be a reference for investigators and scientists working in this field.

A fast implementation of a perfect hash function for picture objects

Abstract In image database systems, symbolic pictures are represented by two-dimensional (2D) strings that are converted into triples. Each triple is mapped to a unique hash address for timely retrieval of pictures, reducing the pattern-matching problem corresponding to a query to that of computation of a hash function. The values associated with the picture objects are used to compute hash addresses for triples developed from the query. Heuristics are proposed to speed up the computation of the associated values for the picture objects. Experimental results show that the new algorithm achieves almost a 90% gain, in search space, over existing algorithms to compute the associated values.

Hierarchical clustering for image databases

The organization of an image database is one of the important issues in efficient storage and retrieval of images. Most of the existing image databases are based on flat structures, with the possibility of an index into the database that can help in narrowing down the images to be searched. In this paper, the author presents a technique to create a hierarchical data structure based on the clustering approach such that a user can select or discard a number of images for subsequent operations. The presented technique is based on application of wavelet analysis to scale the images in hierarchy, and can take advantage of the structure of compressed images in the JPEG 2000 standard

A new approach to knowledge acquisition by repertory grids

Personal construct theory is one of the important tools for knowledge acquisition. It provides an expert’s opinion of the relationship between objects and their properties in the form of a repertory grid. The current techniques provide the repertory grid in the form of a matrix of integers. This paper describes the preliminary attempts for elicitation and analysis of repertory grid using a range of integers for each relationship. Such an elicitation leads to a more natural description of the expert’s knowledge in addition to a better user interface.