Yoda,an adaptive soft classification model: content-based similarity queries and beyond

Abstract. Providing a customized result set based upon a user preference is the ultimate objective of many content-based image retrieval systems. There are two main challenges in meeting this objective: First, there is a gap between the physical characteristics of digital images and the semantic meaning of the images. Secondly, different people may have different perceptions on the same set of images. To address both these challenges, we propose a model, named Yoda, that conceptualizes content-based querying as the task of soft classifying images into classes. These classes can overlap, and their members are different for different users. The “soft” classification is hence performed for each and every image feature, including both physical and semantic features. Subsequently, each image will be ranked based on the weighted aggregation of its classification memberships. The weights are user-dependent, and hence different users would obtain different result sets for the same query. Yoda employs a fuzzy-logic based aggregation function for ranking images. We show that, in addition to some performance benefits, fuzzy aggregation is less sensitive to noise and can support disjunctive queries as compared to weighted-average aggregation used by other content-based image retrieval systems. Finally, since Yoda heavily relies on user-dependent weights (i.e., user profiles) for the aggregation task, we utilize the users' relevance feedback to improve the profiles using genetic algorithms (GA). Our learning mechanism requires fewer user interactions, and results in a faster convergence to the user's preferences as compared to other learning techniques. Correspondence to: Y.-S. Chen (E-mail: yishinc@usc.edu) This research has been funded in part by NSF grants EEC-9529152 (IMSC ERC) and IIS-0082826, NIH-NLM R01-LM07061, DARPA and USAF under agreement nr. F30602-99-1-0524, and unrestricted cash gifts from NCR, Microsoft, and Okawa Foundation.     

A data model and a query languagefor multimedia documents databases

A multimedia application involves information that may be in a form of video, images, audio, text and graphics, need to be stored, retrieved and manipulated in large databases. In this paper, we propose an object-oriented database schema that supports multimedia documents and their temporal, spatial and logical structures. We present a document example and show how the schema can adress all the structures described. We also present a multimedia query specification language that can be used to describe a multimedia content portion to be retrieved from the database. The language provides means by which the user can specify the information on the media as well as the temoral and spatial relationships among these media.     

Toward overcoming culture, skill and situation hurdles in Human-Computer Interaction

This paper proposes a new effective strategy for designing and implementing interactive systems overcoming culture, skill and situation hurdles in Human-Computer Interaction (HCI). The strategy to identify and reduce these hurdles is developed in the framework of a methodology based on a recently introduced model of HCI, and exploits the technological innovations of XML (Extensible Markup Language). HCI is modelled as a cyclic process in which the user and the interactive system communicate by materializing and interpreting a sequence of messages. The interaction process is formalized by specifying both the physical message appearance and the computational aspect of the interaction. This formalization allows the adoption of notation traditionally adopted by users in their workplaces as the starting point of the interactive system design. In this way, the human–system interaction language takes into account the users’ culture. Moreover, the methodology permits user representatives to build a hierarchy of systems progressively adapted to users’ situations, skills and habits, according to the work organization in the domain considered. The strategy is proved to be effective by describing how to implement it using BANCO (Browsing Adaptive Network for Changing user Operativity), a feasibility prototype based on XML, which allows the hierarchy implementation and system adaptations. Several examples from an environmental case under study are used throughout the paper to illustrate the methodology and the effectiveness of the technology adopted. Published online: 4 June 2002     

Integrated spatial and feature image query

We present a new system for querying for images by regions and their spatial and feature attributes. The system enables the user to find the images that contain arrangements of regions similar to those diagrammed in a query image. By indexing the attributes of regions, such as sizes, locations and visual features, a wide variety of complex joint spatial and feature queries are efficiently computed. In order to demonstrate the utility of the system, we develop a process for the extracting color regions from photographic images. We demonstrate that integrated spatial and feature querying using color regions improves image search capabilities over non-spatial content-based image retrieval methods.     

NeTra: A toolbox for navigating large image databases

We present here an implementation of NeTra, a prototype image retrieval system that uses color, texture, shape and spatial location information in segmented image regions to search and retrieve similar regions from the database. A distinguishing aspect of this system is its incorporation of a robust automated image segmentation algorithm that allows object- or region-based search. Image segmentation significantly improves the quality of image retrieval when images contain multiple complex objects. Images are segmented into homogeneous regions at the time of ingest into the database, and image attributes that represent each of these regions are computed. In addition to image segmentation, other important components of the system include an efficient color representation, and indexing of color, texture, and shape features for fast search and retrieval. This representation allows the user to compose interesting queries such as “retrieve all images that contain regions that have the color of object A, texture of object B, shape of object C, and lie in the upper of the image”, where the individual objects could be regions belonging to different images. A Java-based web implementation of NeTra is available at http://vivaldi.ece.ucsb.edu/Netra.     

Drag-and-drop multimedia: ¶an interface framework for digital libraries

In this paper, we describe a new interface for querying multimedia digital libraries and an interface building framework. The interface employs a drag-and-drop style of interaction and combines a structured natural-language style query specification with reusable multimedia objects. We call this interface DanDMM, short for “drag-and-drop multimedia”. DanDMM interfaces capture the syntax of the underlying query language, and dynamically reconfigure to reflect the contents of the data repository. A distinguishing feature of DanDMM is its ability to synthesize integrated interfaces that incorporate both example-based specification using multimedia objects, and traditional techniques including keyword, attribute, and free text-based search. We describe the DanDMM-builder, a framework for synthesizing DanDMM interfaces, and give several examples of interfaces that have been constructed using DanDMM-builder, including a remote-sensing library application and a video digital library. Received: 15 December 1997 / Revised: June 1999     

Classification of document pages using structure-based features

Searching for documents by their type or genre is a natural way to enhance the effectiveness of document retrieval. The layout of a document contains a significant amount of information that can be used to classify it by type in the absence of domain-specific models. Our approach to classification is based on “visual similarity” of layout structure and is implemented by building a supervised classifier, given examples of each class. We use image features such as percentages of text and non-text (graphics, images, tables, and rulings) content regions, column structures, relative point sizes of fonts, density of content area, and statistics of features of connected components which can be derived without class knowledge. In order to obtain class labels for training samples, we conducted a study where subjects ranked document pages with respect to their resemblance to representative page images. Class labels can also be assigned based on known document types, or can be defined by the user. We implemented our classification scheme using decision tree classifiers and self-organizing maps. Received June 15, 2000 / Revised November 15, 2000     

Image retrieval by color semantics

The development of a system supporting querying of image databases by color content tackles a major design choice about properties of colors which are referenced within user queries. On the one hand, low-level properties directly reflect numerical features and concepts tied to the machine representation of color information. On the other hand, high-level properties address concepts such as the perceptual quality of colors and the sensations that they convey. Color-induced sensations include warmth, accordance or contrast, harmony, excitement, depression, anguish, etc. In other words, they refer to the semantics of color usage. In particular, paintings are an example where the message is contained more in the high-level color qualities and spatial arrangements than in the physical properties of colors. Starting from this observation, Johannes Itten introduced a formalism to analyze the use of color in art and the effects that this induces on the user's psyche. In this paper, we present a system which translates the Itten theory into a formal language that expresses the semantics associated with the combination of chromatic properties of color images. The system exploits a competitive learning technique to segment images into regions with homogeneous colors. Fuzzy sets are used to represent low-level region properties such as hue, saturation, luminance, warmth, size and position. A formal language and a set of model-checking rules are implemented to define semantic clauses and verify the degree of truth by which they hold over an image.     

Effectiveness of keyword-based display and selection of retrieval results for interactive searches

We present an approach to increasing the effectiveness of ranked-output retrieval systems that relies on graphical display and user manipulation of “views” of retrieval results, where a view is the subset of retrieved documents that contain a specified subset of query terms. This approach has been implemented in a system named VIEWER (VIEwing WEb Results), acting as an interface to available search engines. An experimental evaluation of the performance of VIEWER in contrast to AltaVista is the major focus of the paper. We first report the results of an experiment on single, short query searches where VIEWER, used as an interactive ranking system, markedly outperformed AltaVista. We then concentrate on a more realistic searching scenario, involving free query formulation, unconstrained selection of retrieval results, and possibility of query reformulation. We report the results of an experiment where the use of VIEWER, compared to AltaVista, seemed to shift the user effort from inspection to evaluation of results, increasing retrieval effectiveness, and user satisfaction. In particular, we found that the VIEWER users retrieved half as many nonrelevant documents as the AltaVista users while retrieving a comparable number of relevant documents. Published online: 22 September 2000     

Spectral covariance and fuzzy regions for image indexing

To improve the discrimination power of color-indexing techniques, we encode a minimal amount of spatial information in the index. We tesselate each image with five partially overlapping, fuzzy regions. In the index, for each region in an image, we store its average color and the covariance matrix of the color distribution. A similiarity function of these color features is used to match query images with images in the database. In addition, we propose two measures to evaluate the performance of image-indexing techniques. We present experimental results using an image database which contains more than 11,600 color images.     

Supporting efficient multimedia database exploration

Due to the fuzziness of query specification and media matching, multimedia retrieval is conducted by way of exploration. It is essential to provide feedback so that users can visualize query reformulation alternatives and database content distribution. Since media matching is an expensive task, another issue is how to efficiently support exploration so that the system is not overloaded by perpetual query reformulation. In this paper, we present a uniform framework to represent statistical information of both semantics and visual metadata for images in the databases. We propose the concept of query verification, which evaluates queries using statistics, and provides users with feedback, including the strictness and reformulation alternatives of each query condition as well as estimated numbers of matches. With query verification, the system increases the efficiency of the multimedia database exploration for both users and the system. Such statistical information is also utilized to support progressive query processing and query relaxation. Received: 9 June 1998/ Accepted: 21 July 2000 Published online: 4 May 2001     

A graphical user interface for Boolean query specification

Online information repositories commonly provide keyword search facilities through textual query languages based on Boolean logic. However, there is evidence to suggest that the syntactic demands of such languages can lead to user errors and adversely affect the time that it takes users to form queries. Users also face difficulties because of the conflict in semantics between AND and OR when used in Boolean logic and English language. Analysis of usage logs for the New Zealand Digital Library (NZDL) show that few Boolean queries contain more than three terms, use of the intersection operator dominates and that query refinement is common. We suggest that graphical query languages, in particular Venn-like diagrams, can alleviate the problems that users experience when forming Boolean expressions with textual languages. A study of the utility of Venn diagrams for query specification indicates that with little or no training users can interpret and form Venn-like diagrams in a consistent manner which accurately correspond to Boolean expressions. We describe VQuery, a Venn-diagram based user interface to the New Zealand Digital Library (NZDL). In a study which compared VQuery with a standard textual Boolean interface, users took significantly longer to form queries and produced more erroneous queries when using VQuery. We discuss the implications of these results and suggest directions for future work. Received: 15 December 1997 / Revised: June 1999     

Approximate query mapping: Accounting for translation closeness

In this paper we present a mechanism for approximately translating Boolean query constraints across heterogeneous information sources. Achieving the best translation is challenging because sources support different constraints for formulating queries, and often these constraints cannot be precisely translated. For instance, a query [score>8] might be “perfectly” translated as [rating>0.8] at some site, but can only be approximated as [grade=A] at another. Unlike other work, our general framework adopts a customizable “closeness” metric for the translation that combines both precision and recall. Our results show that for query translation we need to handle interdependencies among both query conjuncts as well as disjuncts. As the basis, we identify the essential requirements of a rule system for users to encode the mappings for atomic semantic units. Our algorithm then translates complex queries by rewriting them in terms of the semantic units. We show that, under practical assumptions, our algorithm generates the best approximate translations with respect to the closeness metric of choice. We also present a case study to show how our technique may be applied in practice. Received: 15 October 2000 / Accepted: 15 April 2001 Published online: 28 June 2001     

Views in a large-scale XML repository

We are interested in defining and querying views in a huge and highly heterogeneous XML repository (Web scale). In this context, view definitions are very large, involving lots of sources, and there is no apparent limitation to their size. This raises interesting problems that we address in the paper: (i) how to distribute views over several machines without having a negative impact on the query translation process; (ii) how to quickly select the relevant part of a view given a query; (iii) how to minimize the cost of communicating potentially large queries to the machines where they will be evaluated. The solution that we propose is based on a simple view definition language that allows for automatic generation of views. The language maps paths in the view abstract DTD to paths in the concrete source DTDs. It enables a distributed implementation of the view system that is scalable both in terms of data and load. In particular, the query translation algorithm is shown to have a good (linear) complexity. Received: November 1, 2001 / Accepted: March 2, 2002 Published online: September 25, 2002     

Extra-Ordinary Human–Machine Interaction: What can be Learned from People with Disabilities?

When specifying, designing and implementing technology for work, it is very important to consider the whole of the potential user group, and not to forget that this might include people with disabilities. Obvious reasons for this are demography, disabled people forming a significant and growing proportion of the workforce, and legal obligations caused by equal opportunities legislation. There are more subtle reasons, however, which link so-called ‘disabled people’ with the rest of the population. The paper notes that most people have some disabilities, sometimes temporarily due to accident, or permanently due to ageing and other processes. In addition it notes that even otherwise fully functioning people can be seriously handicapped by hostile environments, and a parallel is drawn between the human interface needs of disabled people operating in ordinary situations and able-bodied people operating in high workload and stress situations. The paper concludes by suggesting a research agenda for the field on the basis of these considerations.     

Interactive authoring of multimedia documents in a constraint-based authoring system

As multimedia applications spread widely, it is crucial for programming and design support systems to handle “time” in multimedia documents effectively and flexibly. This paper presents a set of interactive system support tools for designing and maintaining the temporal behavior of multimedia documents. The tool set provides mechanisms for anomaly detection, temporal query processing, and interactive scheduling. It is based on a fast incremental constraint solver we have developed, which can be adapted by any constraint-based system. The incremental constraint solver provides immediate feedback to the user, supporting a highly interactive design process. Combined with existing optimal layout generation mechanisms proposed in the literature, our tools effectively utilize the flexibility provided by constraint-based systems.     

The end of zero-hit queries: query previews for NASA’s Global Change Master Directory

The Human-Computer Interaction Laboratory (HCIL) of the University of Maryland and NASA have collaborated over three years to refine and apply user interface research concepts developed at HCIL in order to improve the usability of NASA data services. The research focused on dynamic query user interfaces, visualization, and overview + preview designs. An operational prototype, using query previews, was implemented with NASA’s Global Change Master Directory (GCMD), a directory service for earth science datasets. Users can see the histogram of the data distribution over several attributes and choose among attribute values. A result bar shows the cardinality of the result set, thereby preventing users from submitting queries that would have zero hits. Our experience confirmed the importance of metadata accuracy and completeness. The query preview interfaces make visible the problems or gaps in the metadata that are undetectable with classic form fill-in interfaces. This could be seen as a problem, but we think that it will have a long-term beneficial effect on the quality of the metadata as data providers will be compelled to produce more complete and accurate metadata. The adaptation of the research prototype to the NASA data required revised data structures and algorithms. Received: 12 December 1997 / Revised: June 1999     

Integrating symbolic images into a multimedia database system using classification and abstraction approaches

Symbolic images are composed of a finite set of symbols that have a semantic meaning. Examples of symbolic images include maps (where the semantic meaning of the symbols is given in the legend), engineering drawings, and floor plans. Two approaches for supporting queries on symbolic-image databases that are based on image content are studied. The classification approach preprocesses all symbolic images and attaches a semantic classification and an associated certainty factor to each object that it finds in the image. The abstraction approach describes each object in the symbolic image by using a vector consisting of the values of some of its features (e.g., shape, genus, etc.). The approaches differ in the way in which responses to queries are computed. In the classification approach, images are retrieved on the basis of whether or not they contain objects that have the same classification as the objects in the query. On the other hand, in the abstraction approach, retrieval is on the basis of similarity of feature vector values of these objects. Methods of integrating these two approaches into a relational multimedia database management system so that symbolic images can be stored and retrieved based on their content are described. Schema definitions and indices that support query specifications involving spatial as well as contextual constraints are presented. Spatial constraints may be based on both locational information (e.g., distance) and relational information (e.g., north of). Different strategies for image retrieval for a number of typical queries using these approaches are described. Estimated costs are derived for these strategies. Results are reported of a comparative study of the two approaches in terms of image insertion time, storage space, retrieval accuracy, and retrieval time. Received June 12, 1998 / Accepted October 13, 1998     

Progressive evaluation of nested aggregate queries

In many decision-making scenarios, decision makers require rapid feedback to their queries, which typically involve aggregates. The traditional blocking execution model can no longer meet the demands of these users. One promising approach in the literature, called online aggregation, evaluates an aggregation query progressively as follows: as soon as certain data have been evaluated, approximate answers are produced with their respective running confidence intervals; as more data are examined, the answers and their corresponding running confidence intervals are refined. In this paper, we extend this approach to handle nested queries with aggregates (i.e., at least one inner query block is an aggregate query) by providing users with (approximate) answers progressively as the inner aggregation query blocks are evaluated. We address the new issues pose by nested queries. In particular, the answer space begins with a superset of the final answers and is refined as the aggregates from the inner query blocks are refined. For the intermediary answers to be meaningful, they have to be interpreted with the aggregates from the inner queries. We also propose a multi-threaded model in evaluating such queries: each query block is assigned to a thread, and the threads can be evaluated concurrently and independently. The time slice across the threads is nondeterministic in the sense that the user controls the relative rate at which these subqueries are being evaluated. For enumerative nested queries, we propose a priority-based evaluation strategy to present answers that are certainly in the final answer space first, before presenting those whose validity may be affected as the inner query aggregates are refined. We implemented a prototype system using Java and evaluated our system. Results for nested queries with a level and multiple levels of nesting are reported. Our results show the effectiveness of the proposed mechanisms in providing progressive feedback that reduces the initial waiting time of users significantly without sacrificing the quality of the answers. Received April 25, 2000 / Accepted June 27, 2000