Catálogo de publicaciones - libros

To overcome the drawbacks of both keywords-based and content-based image retrieval approaches, we propose a hybrid solution for image retrieval on WWW. A content organisation system is introduced to assess content-based similarity of results returned from contemporary image search engines so that the results are organised with content similarity, and alternative search modes such as query by example are also enabled.

We present a clustering approach for video summarization and browsing based on the Growing Cell Structures (GCS) neural algorithms and MPEG-7 video descriptors. Each I frame is represented as a feature vector of MPEG-7 descriptors. These vectors are clustered using GCS to select one or more keyframes for each shot. The extracted keyframes are then grouped together by TreeGCS to form a hierarchical view of the video for efficient browsing. We evaluate the effectiveness of different combinations of MPEG-7 descriptors and also compare them with the performance of color histograms.

Much of the world’s data is in the form of time series, and many other types of data, such as video, image, and handwriting, can easily be transformed into time series. This fact has fueled enormous interest in time series retrieval in the database and data mining community. However, much of this work’s narrow focus on efficiency and scalability has come at the cost of usability and effectiveness. Here, we introduce a general framework that learns a distance measure with arbitrary constraints on the warping path of the Dynamic Time Warping calculation. We demonstrate utility of our approach on both classification and query retrieval tasks for time series and other types of multimedia data, then show that its incorporating into the relevance feedback system and query refinement can further improve the precision/recall by a wide margin.

There has been a huge increase in the utilization of video as one of the most preferred type of media due to its content richness for many significant applications including sports. To sustain an ongoing rapid growth of sports video, there is an emerging demand for a sophisticated content-based indexing system. Users recall video contents in a high-level abstraction while video is generally stored as an arbitrary sequence of audio-visual tracks. To bridge this gap, this paper will demonstrate the use of domain knowledge and characteristics to design the extraction of high-level concepts directly from audio-visual features. In particular, we propose a multi-level semantic analysis framework to optimize the sharing of domain characteristics.

The MPEG-7 texture description tools are applied to the retrieval of the historical aerial photographs. These aerial photographs had been digitized into gray-scaled and low-resolution images to build a digital library. The aerial image retrieval over the digital library is achieved on the basis of 1) texture feature extraction methods, which are homogeneous texture, texture browsing and edge histogram in the MPEG-7 texture descriptors, and 2) their corresponding similarity measurements. The image retrieval methods are evaluated over a collection of historical aerial photographs. The paper presents the retrieval results of these texture feature extraction methods.

Inducing general functions from specific training examples is a central problem in the machine learning. Using sets of If-then rules is the most expressive and readable manner. To find If-then rules, many induction algorithms such as ID3, AQ, CN2 and their variants, were proposed. Sequential covering is the kernel technique of them. To avoid testing all possible selectors, Entropy gain is used to select the best attribute in ID3. Constraint of the size of star was introduced in AQ and beam search was adopted in CN2. These methods speed up their induction algorithms but many good selectors are filtered out. In this work, we introduce a new induction algorithm that is based on enumeration of all possible selectors. Contrary to the previous works, we use pruning power to reduce irrelative selectors. But we can guarantee that no good selectors are filtered out. Comparing with other techniques, the experiment results demonstrate that the rules produced by our induction algorithm have high consistency and simplicity.

Machine learning methods such as fuzzy logic, neural networks and decision tree induction have been applied to learn rules, however they can get trapped into a local optimal. Based on the principle of natural evolution and global searching, a genetic algorithm is promising for obtaining better results. This article adopts the learning classifier systems (LCS) technique to provide a hybrid knowledge integration strategy, which makes for continuous and instant learning while integrating multiple rule sets into a centralized knowledge base. This paper makes three important contributions: (1) it provides a knowledge encoding methodology to represent various rule sets that are derived from different sources, and that are encoded as a fixed-length bit string; (2) it proposes a knowledge integration methodology to apply genetic operations and credit assignment to generate optimal rule sets; (3) it uses three criteria (accuracy, coverage, and fitness) to apply the knowledge extraction process, which is very effective in selecting an optimal set of rules from a large population. The experiments prove that the rule sets derived by the proposed approach is more accurate than the Fuzzy ID3 algorithm.

After the doubtful success of content-based e-learning systems, simulations are gaining momentum within the e-learning community. Along this line we are working on JVM, a simulation-based learning environment to teach the Java Virtual Machine (JVM) and the compilation of object-oriented languages. This paper describes both the metaphorical simulation of the JVM and the knowledge our system possesses and details an execution example that reflects how all the information is used on it.

This study proposes an idiotype network model system adopted an “asymmetric” idiotype – anti-idiotypic interaction, and shows realizing immune tolerance as well as immune memory on a transient dynamics. To date, the immune tolerance was a problematic phenomenon for traditional idiotype network models adopted “symmetric” idiotypic interaction because they failed to reproduce it. This paper reports the proposed model succeeds in the display of the immune tolerance by considering an asymmetric anti-idiotypic interaction. Actually, a computational experiment clarifies that its establishment associates with high anti-idiotypic idiotype’s population level of when antigens are re-dosed. This result indicates what the anti-idiotypic idiotype functions effectively plays a decisive role for the establishment of the immune tolerance. Lastly, this study closes at the suggestion that the asymmetric modeling is more immunologically relevant than the symmetric one which most theoreticians supported so far.

In our previous research, we proposed a new approach for mutual repairing system using the immunity-based diagnostic mobile agents. However, we have not yet analyzed the performance by more detailed simulations and/or mathematical models. In this paper, the approach using the immunity-based diagnostic mobile agent is compared with other approaches using majority vote model or host-to-host communication. Some results show that the immunity-based diagnostic mobile agents can repair more abnormal hosts than the other methods. We also address a mathematical model to explain a phase transition.