We introduce an image database called Natural Image Database of collecting natural color images that can be used via the Internet, and then show how it is used for illuminant classification and evaluate our sensor correlation method. A significant feature of the present database is inclusion of both image and illuminant data. It is not only a collection of color images taken a CCD camera, but also includes the calibration data of the camera used for image acquisition and the illuminant spectral data measured by a spectro-radiometer for each scene. Therefore, illuminant estimation algorithms can be evaluated on various natural scenes and comparison with the direct measurements. The illuminant classification is based on correlation computation between the gamut of an observed image and each of the reference illuminant gamuts. The illuminant gamuts of blackbody radiators are created from 2500K to 8500K in the RB sensor plane. The image gamut is defined as the convex hull of pixel values of the observed image. Then, one illuminant gamut is selected as the estimated illuminant, which gives an overall maximum of the correlation function. An experiment using the image database is executed for demonstrating the feasibility.

Real-time interactive slicing is a tool of choice for exploring 3D anatomic data sets such as the Visible Human. We offer real-time slicing on the Web by partitioning the application between a Java applet (the client) and the Web server. Two approaches for partitioning the work between the client and the server are presented and compared. In the first approach, we transfer complete compressed slices from the server to the client. In the second approach, we successively build a local cache on the client by transferring small subvolumes of increasing resolution from the server to the client. The client is responsible for extracting the displayed slices from the local data cache. The Web-based Real-time Visible Human navigator can be accessed at our Visible Human Web site at http://visiblehuman.epfl.ch. A high-bandwidth network connection is recommended. The web server offering real-time interactive slicing together with anatomic structure labeling opens new perspectives for teaching anatomy to paramedical and medical professions.

We have designed and implemented a 3D digital atlas of zebrafish development. In this paper we focus on the presentation of the data of this atlas using the Internet. An important part of the data of this atlas is graphical. Therefore special effort is put into fast and comprehensive visualization of the 2D and 3D graphical data, i.e. images and their annotations as well as the 3D models that arise from the application of the annotation process on a coherent 3D dataset. These data are presented to the end-user and the modes of visualization need be linked to provide the user with feedback. The success of visualization over the Internet is partially depending on the transfer time of the data as well as the computational load of the data at the client side. The applications discussed in this paper are optimized for transfer time and computational load.

The paper is dealing with the newest developments in PDF standardization especially for imaging in Electronic Publishing and Internet publishing. Since the mid-1980s, page production has shifted from relatively craft-based industry a wide-open consumer-based phenomenon. PDF/X can be used as a shorthand way of specifying most of what you need to tell somebody in order for them to create a file which can be printed or published correctly when they send it from one place to another.

In this paper, we propose a contents design guideline for mobile Internet terminals (browser phones) such as i-mode phones. This guideline shows specs of the browser phones in Japan, and considers situations and needs of common users. Then we define design rules.

Internet use by the people with disabilities and the elderly in Japan is still low, but growing. However, the majority of web contents written in Japanese, even government sites, have very low accessibility. This paper introduces the active measures being taken in Japan to improve such conditions; consideration of a web contents accessibility guideline tailored to the unique characteristics of the Japanese language, development of a system to evaluate accessibility and implementation of actual trials.

Recently with the rapid growth of multimedia industry, images are used by many people for a variety of applications. To represent and retrieve these images efficiently, various methods have been proposed. In this paper, A new method for image retrieval is presented. This method uses multiple descriptions at different levels to represent images. This model allows the use of all levels of descriptions in the retrieval process, hence permits not only semantic search but also syntactic search. The quick content-based visual access to the stored images is essential for efficient navigation through image collections. At the time of visual access to image database, the users have very different objectives. Users' interests may vary from one person to another. In order to satisfy the information needs of users, it is of paramount importance to effectively and efficiently adapt the retrieval process to each user. In our approach the retrieval process is guided by the user's retrieval intentions according to his/her needs and preferences. These characteristics of the retrieval method increase the retrieval efficiency and makes the model very flexible as it can be used universally for retrieving images from different domains. We have applied the proposed approach to a range of color images, and obtained positive results.

This paper proposes a new style tool, STAMP (Spatio-Temporal Associations with Multiple Photographs), to create new style of pseudo 3-D spaces. This system allows ordinary people to create, publish, share and navigate pseudo 3-D spaces easily using perspective photos on the Web. We present a framework of STAMP and functions of our developed prototype system using the framework. Some experiments with STAMP are also reported.

This paper describes how a query system can exploit the basic knowledge by employing semi-automatic relevance feedback to refine queries and runtimes. For general databases, it is often useless to call complex attributes, because we have not sufficient information about images in the database. Moreover, these images can be topologically very different from one to each other and an attribute that is powerful for a database category may be very powerless for the other categories. The idea is to use very simple features, such as color histogram, correlograms, Color Coherence Vectors (CCV), to fill out the signature vector. Then, a number of mixture vectors is prepared depending on the number of very distinctive categories in the database. Knowing that a mixture vector is a vector containing the weight of each attribute that will be used to compute a similarity distance. We post a query in the database using successively all the mixture vectors defined previously. We retain then the N first images for each vector in order to make a mapping using the following information: Is image I present in several mixture vectors results? What is its rank in the results? These informations allow us to switch the system on an unsupervised relevance feedback or user's feedback (supervised feedback).

There is an increasing interest in the computer animation. The most of the current animation software tools are described by Richard Parent in [5]. The driving ideas behind animation in general and computer animation in particular are presented in [6]. More and more animations are available in various places on the Internet. The increasing availability of potentially interesting materials makes the search extremely difficult especially when the search is by the animation contents. This work is devoted to analyzing the syntax contents of the animations. In the paper we present the ANIMATION system - a system for animation scene and contents creation, retrieval and display. The system is based on MPEG-4 standard [3, 4]. MPEG-4 is an ISO/IEC standard developed by MPEG (Moving Picture Experts Group). These standard makes interactive video on CD-ROM and Digital Television possible.

In this paper we present a three-dimensional model retrieval system. A three-dimensional model is described by two invariant descriptors : a shape index and a histogram of distances between meshes. This work focuses on extracting invariant descriptors that well represent a three-dimensional model, and on combining theses descriptors in order to get a better retrieval performance. An experimental evaluation demonstrates the good performance of the approach.

A new television system based on the Autosophy information theory is now being developed for Internet video applications. Autosophy communication is already widely used on the Internet, with current applications including data compression in V.42bis modems, and the lossless still image compression standards GIF and TIF. Now Autosophy Internet video is being developed and is ready for demonstration. In conventional television, bit rates are determined entirely by screen size, resolution, and scanning rates. The images shown on the screen are irrelevant, such that random noise video requires the same bit rate as any other video content. In the new Autosophy-based television, in contrast, bit rates are determined entirely by the video content, essentially motion and complexity within the images. A very high degree of visually lossless video compression is possible because only moving portions of the video are transmitted. Transmitted codes represent multi-pixel image clusters found in a pre-grown hyperspace library. The system can dynamically reduce resolution of fast-moving objects when necessary to accommodate bandwidth restrictions. Ideally suited to the Internet environment, the new television also features high resistance to delayed or dropped packets, a universal hardware-independent communication format, and optional codebook encryption for secure communications.

This paper presents a data model for images immersed in the world wide web and that derive their meaning from visual similarity, from the connection with the text of the pages that contains them, and from the link structure of the web. I will model images on the web as a graph whose nodes are either text documents or images, and whose edges are links, labeled with measures of relevance of one document towards the other. The paper presents briefly the features used to characterize the text and the visual aspect of the images, and then goes on to present a data algebra suitable to navigate and query the database.

Most indexing systems are based on global image descriptors. Nowadays, new representations are used. They try to describe images more precisely without exploiting any semantic description. They use local statistics and their relationships in the image. In this paper, we present these approaches and introduce a new representation system based on a pyramidal graph. First results are also presented which show that the proposed system is very promising for both partial and global requests.

The purpose of our visual information retrieval tool is to extract from a database images that are similar to an image query. Color features are generally used to define a measure of similarity between images, as they are usually very robust to noise, image degradation, changes in size, resolution or orientation. Nevertheless, the most often features suffer objectively from the lack of color spatial knowledge. Then, our purpose is to merge two classical methods : the color pyramid and the interest points detection, well-known for grey level image analysis. The pertinence of this new method is demonstrated by an evaluation and a comparison with others keypoints detectors. We show the interest for image indexation with concrete tests on our large images database, using the icobra system.

This paper deals with Computer Aided Diagnosis for skin cancers (melanomas). The diagnosis is based on some rules called the ABCD mnemonics. They take into account color distribution, lesion's diameter, etc. The goal isn't to classify the lesion but to find those, which are the most similar, in order to help the expert to confirm his diagnosis and to avoid any useless excision. This is done thanks to an indexation system, which compare the signatures of previously diagnosed lesions contained in a database and patient's lesion signature. This last is constructed by translating the rules into image processing attributes. We have divided then into three families: Color attributes (color and fuzzy histograms), Texture attributes (co-occurrence matrix and Haralick indices) and Shape attributes (lesion surface and maximum included circle). Image quantization permits us to keep only the most significant colors, thus giving a light structure. Finally, we define a distance for each attribute and use weighted combination for the similarity measure.

Through a state funded grant initiative, the Department of Art at the University of Idaho has recently founded an information visualization studio. We specialize in interdisciplinary research on developing visualization techniques that allow a transformation of complex data into visual formats while providing clarity and exposing underlying structures. Clear and concise visualization of information makes this understanding more immediate. It also allows for added critical thinking about a subject. For example, in the area of hardware and software design, existing visualizations of structures are often clouded by an engineer's previous knowledge about the subject. They know what they are looking at, but a machine doesn't, nor does an end user. Clear graphics developed by an information designer make it easier to see problems and give parameters that a programmer can then apply to building a better system. We seek to produce visualization systems, methodologies, and examples for the resolution of problems where visual demonstration of concepts and materials enhances, even permits, understanding. We intend to summarize the findings of our project by illustrating and demonstrating techniques, which we have developed, for manipulating raw data into clear graphic presentations of complex systems for decision making. We will also summarize our findings on methods for creating visualizations designed to assist individuals with perceptual or cognitive difficulties.

The paper describes a methodology we use to translate an existing conventional archive into a digital one. The method does well for large archives comprising documents with essential graphic constituent (handwritten texts, photographs, drawings, etc.). Main structural components of our digital archive are relational database and image bank which are physically separated but logically linked together. The components make up three-level distributed structure consisting of primary archive, its regional replicas, and various secondary archives (among them subsets presented in the Web and collections of compact discs). Only authorizes user are allowed to access two upper levels, and the bottom level is open for free public access. A secondary archive is created and updated automatically without special development. Such construction allows us to combine reliable storage, easy access and protection of intellectual property. The paper also presents several digital archives already implemented in the Archive of the Russian Academy of Sciences.

With the growth of Internet and storage capability in recent years, image has become a widespread information format in World Wide Web. However, it has become increasingly harder to search for images of interest, and effective image search engine for the WWW needs to be developed. We propose in this paper a selective filtering process and a novel approach for image classification based on feature element in the image search engine we developed for the WWW. First a selective filtering process is embedded in a general web crawler to filter out the meaningless images with GIF format. Two parameters that can be obtained easily are used in the filtering process. Our classification approach first extract feature elements from images instead of feature vectors. Compared with feature vectors, feature elements can better capture visual meanings of the image according to subjective perception of human beings. Different from traditional image classification method, our classification approach based on feature element doesn't calculate the distance between two vectors in the feature space, while trying to find associations between feature element and class attribute of the image. Experiments are presented to show the efficiency of the proposed approach.

Content-based image and, more generally, multimedia retrieval calls for a semantic understanding of the content of an image. However, the discrepancy between any automated segmentation technique and a operator-based segmentation is now established. There is therefore a strong need for a technique that exploits the best of analytical segmentation techniques while remaining flexible enough to allow for a useful decomposition of the image. In this paper, we describe an interactive image segmentation framework and applications where this framework is essential. Our technique is general enough to allow a generic segmentation technique to be embedded and optimally exploited. The principle is to efficiently obtain a fine-to-coarse image decomposition and represent this structure using an XML description scheme. When browsing through this pyramid, the user will retrieve and mark semantically useful structures within the document. The result may then be used in different contexts such as image indexing and annotation, object-based multimedia query formulation, object tracking in video, etc. We first study the conditions for a base segmentation technique to fit within our framework. Then, we show that our framework provides a solution to typical problems encountered in the context of multimedia information indexing and retrieval. We detail an example Java-based implementation which may be used as a stand-alone tool or as a module within other applications.

The need for a general and comprehensive vocabulary for annotating images is described. Current image indexing vocabularies and their problems are discussed, and research into image features described by humans in various tasks is reviewed. The development of an image indexing vocabulary id described, and a conceptual structure which relates the features of the vocabulary to the requirements of image annotation is presented. The process for building and testing the vocabulary as a tool in benchmarking is then discussed, as are the research questions involved.

Annotating image collections is crucial for different multimedia applications. Not only this provides an alternative access to visual information but it is a critical step to perform the evaluation of content-based image retrieval systems. Annotation is a tedious task so that there is a real need for developing tools that lighten the work of annotators. The tool should be flexible and offer customization so as to make the annotator the most comfortable. It should also automate the most tasks as possible. In this paper, we present a still image annotation tool that has been developed with the aim of being flexible and adaptive. The principle is to create a set of dynamic web pages that are an interface to a SQL database. The keyword set is fixed and every image receives from concurrent annotators a set of keywords along with time stamps and annotator Ids. Each annotator has the possibility of going back and forth within the collection and its previous annotations. He is helped by a number of search services and customization options. An administrative section allows the supervisor to control the parameter of the annotation, including the keyword set, given via an XML structure. The architecture of the tool is made flexible so as to accommodate further options through its development.

Traditional content-based image retrieval (CBIR) systems suffer from at least two shortcomings. First, most systems require users to provide good images to initiate queries. We argue that finding good seeds is the job of the search engine itself, and this circular requirement leaves the core problem---understanding users' query concepts---unsolved. The second shortcoming is that most systems fail to adequately model users' subjectivity. This paper proposes two test queries to Benchathlon for steering the CBIR community to seriously address the above issues. We also present a search scenario to show how such queries may be supported and their performance be measured.

In this paper we present an extension to Quicklink that allows the user to browse database contents according to criteria regarding the semantics of the objects represented in the images. The Quicklink system retrieves images similar to a query image from large archives of artworks by dynamically matching their accompanying textual descriptions, and presents the results in HTML pages, where the images retrieved are ordered according to their degree of similarity. It is designed to adapt its behavior to user requests through a relevance feedback mechanism. The core of the extension designed and implemented in Quicklink consists in the Visual Dictionary, described here.

We present here a web-based protytpe for the interactive search of items in quality electronic catalogues. The system based on a multimedia information retrieval architecture, allows the user to query a multimedia database according to several retrieval strategies, and progressively refine the system's response by indicating the relevance, or non-relevance of the items retrieved. Once a subset of images meeting the user's information needs have been identified, these images can be displayed in a virtual exhibition that can be visited interactively by the user exploiting VRML technology.

This paper introduces a system for image retrieval by examples based on user warpable image similarity metrics. The system supports positive and negative examples with relevance feedback, providing an approximation to fuzzy Boolean query specification.

TV News is a well-structured media, since it has distinct boundaries of semantic units (news stories) and relatively constant content structure. Hence, an efficient algorithm to segment and analyze the structure information among news videos would be necessary for indexing or retrieving a large video database. Lots of researches in this area have been done by using close-caption, speech recognition or Video-OCR to obtain the semantic content, however, these methods put much emphasis on obtaining the text and NLP for semantic understanding. Here, in this paper, we try to solve the problem by integrating statistic model and visual features. First, a video caption and anchorperson shot detection method is presented, after that, a statistic model is used to describe the relationship between the captions and the news story boundaries, then, a news story segmentation method is introduced by integrating all these aforementioned results. The experiment results have proved that the method can be used in acquiring most of the structure information in News programs.

The strategy for a self-consistent procedure for deriving invariance preserving cost functions using the Extreme Physical Information theory is outlined. The self-adjoint invariant EPI solutions are obtained by minimizing the constrained EPI cost function is with respect to the Lie derivative (Noether's theorem). The undetermined quantities are specified in a manner so as to satisfy the Helmholtz condition. To provide for an example, the case of model-based interpolation of high dimensional data, approximated by clusters is discussed.

The paper describes a method for detecting a color cast (i.e. a superimposed dominant color) in a digital image without any a priori knowledge of its semantic content. The color gamut of the image is first mapped in the CIELab color space. The color distribution of the whole image and of the so-called Near Neutral Objects (NNO) is then investigated using statistical tools then, to determine the presence of a cast. The boundaries of the near neutral objects in the color space are set adaptively by the algorithm on the basis of a preliminary analysis of the image color gamut. The method we propose has been tuned and successfully tested on a large data set of images, downloaded from personal web-pages or acquired using various digital and traditional cameras.

There are a number of color encoding standards that have lately been developed ant are being proposed to the imaging community for different encoding steps in the color imaging workflow, optimized for different purposes. In this paper, their applicability for master and derivative image file encoding for image databases is discussed. Several types of image databases are defined, and the advantages and disadvantages of archiving master files after different image workflow steps are outlined. Several input- and output-referred encodings are summarized and discussed in relation to their applicability for image database file encoding.