Ngoc Binh Nguyen

Nonhierarchical Document Clustering Based on a Tolerance Rough Set Model

Document clustering, the grouping of documents into several clusters, has been recognized as a means for improving efficiency and effectiveness of information retrieval and text mining. With the growing importance of electronic media for storing and exchanging large textual databases, document clustering becomes more significant. Hierarchical document clustering methods, having a dominant role in document clustering, seem inadequate for large document databases as the time and space requirements are typically of order O(N3) and O(N2), where N is the number of index terms in a database. In addition, when each document is characterized by only several terms or keywords, clustering algorithms often produce poor results as most similarity measures yield many zero values. In this article we introduce a nonhierarchical document clustering algorithm based on a proposed tolerance rough set model (TRSM). This algorithm contributes two considerable features: (1) it can be applied to large document databases, as the time and space requirements are of order O(NlogN) and O(N), respectively; and (2) it can be well adapted to documents characterized by a few terms due to the TRSMs ability of semantic calculation. The algorithm has been evaluated and validated by experiments on test collections.

Hierarchical Document Clustering Based on Tolerance Rough Set Model

Clustering is a powerful tool for knowledge discovery in text collections. The quality of document clustering depends not only on clustering algorithms but also on document representation models. We develop a hierarchical document clustering algorithm based on a tolerance rough set model (TRSM) for representing documents, which offers a way of considering semantics relatedness between documents. The results of validation and evaluation of this method suggest that this clustering algorithm can be well adapted to text mining.

Documents clustering using tolerance rough set model and its application to information retrieval

Clustering is a powerful tool for analyzing and finding useful information in text collections. However, document clustering is a difficult clustering problem because of the unstructured form and textual characteristics of documents. As a consequence, the quality of document clustering depends not only on clustering algorithms but also on document representation models. In this work we introduce a tolerance rough set model (TRSM) for representing documents as an alternative way of considering semantics relatedness between documents. Using TRSM we develop two hierarchical and nonhierarchical clustering algorithms for documents and apply these clustering methods to information retrieval. The TRSM clustering methods and the TRSM cluster-based information retrieval method are carefully evaluated and validated by comparative experiments on test collections.

Shot Type and Replay Detection for Soccer Video Parsing

Parsing the structure of soccer video plays an important role in semantic analysis of soccer video. In this paper, we present a shot classification method based on the detection of grass field pixels and size of players. In addition, a replay detection algorithm is proposed. First, the candidate logo images are identified by using contrast feature and histogram difference. The contrast logo template is calculated to detect logo frames. Finally, replay segments are identified by pairing and finding the beginning and the end of logo transition. Experiments on three soccer matches showed that our method is effective and applicable for higher level semantic analysis.

Soccer video summarization based on cinematography and motion analysis

Summarization of soccer videos has been widely studied due to its worldwide viewers and potential commercial applications. Most existing methods focus on searching for highlight events in soccer videos such as goals, penalty kicks and generating a summary as a list of such events. However, besides highlight events, scenes of intensive competition between players of two teams and emotional moments are also interesting. In this paper, we propose a soccer summarization system which is able to capture highlight events, scenes of intensive competition, and emotional moments. Based on the flow of soccer games, we organize a video summary as follows: first, scenes of intensive competition, second, what events happened, third, who were involved in the events, and finally how players or audience reacted to the events. With this structure, the generated summary is more complete and interesting because it provides both game play and emotional moments. Our system takes broadcast video as input, and divides it into multiple clips based on cinematographic features such as sport video production techniques, the transition of shots, and camera motions. Then, the system evaluates the interest level of each clip to generate a summary. Experimental results and subjective evaluation are carried out to evaluate the quality of the generated summary and the effectiveness of our proposed interest level measure.

A Mixed Similarity Measure in Near-Linear Computational Complexity for Distance-Based Methods

Many methods of knowledge discovery and data mining are distance-based such as nearest neighbor classification or clustering where similarity measures between objects play an essential role. While real-world databases are often heterogeneous with mixed numeric and symbolic attributes, most available similarity measures can only be applied to either symbolic or numeric data. In such cases, data mining methods often require transforming numeric data into symbolic ones by discretization techniques. Mixed similarity measures (MSMs) without discretization of numeric values are desirable alternatives for objects with mixed symbolic and numeric data. However, the time and space complexities of computing available MSMs are often very high that make MSMs not applicable to large datasets. In the framework of Goodalls MSM inspired by biological taxonomy, computing methods have been done but their time and space complexities so far are at least O(n2 log n2) and O(n2), respectively. In this work, we propose a new and efficient method for computing this MSM with O(n log n) time and O(n) space complexities. We demonstrate experimentally the applicability of new method to large datasets and suggest meta-knowledge on the use of this MSM. Practically, the experimental results show that only the near-linear time and space MSM could be applicable to mining large heterogeneous datasets.

Human Interaction Recognition Using Independent Subspace Analysis Algorithm

Human interaction recognition has been widely studied because it has great scientific importance and many practical applications. Most existing methods rely on spatio-temporal local features (i.e. SIFT), human poses, and human joints to model human interactions. Motivated by the success of deep learning networks, we introduce a three-layer convolutional network which uses the Independent Subspace Analysis (ISA) algorithm to learn hierarchical invariant features from videos. The obtained invariant features are used as the inputs to a standard bag-of-features (BOF) model to recognize human interactions. We investigate the performance of our approach and the effectiveness of hierarchical invariant features on video sequences of the UT-Interaction dataset which contain both interacting persons and irrelevant pedestrians in the scenes. Experimental results show that our three-layer convolutional ISA network is able to learn features which are effective to represent complex activities such as human interactions in realistic environments.

Cluster-based Information Retrieval with Tolerance Rough Set Model

The objectives of this paper are twofold. First is to introduce a model for representing documents with semantics relatedness using rough sets but with tolerance relations instead of equivalence relations (TRSM). Second is to introduce two document hierarchical and nonhierarchical clustering algorithms based on this model and TRSM cluster-based information retrieval using these two algorithms. The experimental results show that TRSM offers an alterative approach to text clustering and information retrieval.

Classification and Temporal Localization for Human-Human Interactions

Recognition of human-human interactions is one of the most important topics since it has great scientific importance and many potential practical applications such as surveillance, and automatic video indexing. Previous approaches have only concentrated on classification and put less effort into localization of human interactions. In addition, they rely on hand-designed features (e.g. SIFT, HOG), or human poses or human joints to model human interactions. A disadvantage of such approaches is that it is difficult and time consuming to extend these features to different datasets in the real world. In this paper, we approach the problem of human interaction classification and temporal localization with unsupervised feature learning. Motivated by the well-known Independent Subspace Analysis (ISA) in natural image statistics and the convolution technique, we introduce a three-layer convolutional ISA network to learn hierarchical invariant features from videos. Using the invariant features learned by the three-layer convolutional ISA network, we build a bag-of-features (BOF) representation for videos. We then apply Support Vector Machine (SVM) to classify human interactions, and employ a sliding window technique to localize interactions temporally. We also evaluate the performance of classification and temporal localization on video sequences of the UT-Interaction dataset and Hollywood dataset. The encouraging results on classification show that our three-layer convolutional ISA network is able to learn features which are effective to represent complex activities such as human interactions in realistic environments. Although temporal localization results are insufficient for real applications, it is a first step for further research in localization of human interactions.

Human Interaction Recognition Using Hierarchical Invariant Features

Human interaction recognition has been widely studied because it has great scientific importance and many potential practical applications. However, recognizing human interactions is also very challenging especially in realistic environments where the background is dynamic or has varying lighting conditions. Most existing methods rely on either spatio-temporal local features (i.e. SIFT) or human poses, or human joints to model human interactions. As a result, they are not fully unsupervised processes because they require either hand-designed features or human detection results. Motivated by the recent success of deep learning networks, we investigate a three-layer convolutional network which uses the Independent Subspace Analysis (ISA) algorithm to learn hierarchical invariant features from videos. Using the invariant features learned by the ISA, we build a bag-of-features (BOF) model to recognize human interactions. We also evaluate the performance of our approach and the effectiveness of hierarchical invariant features on video sequences of the UT-Interaction dataset which contain both interacting persons and irrelevant pedestrians in the scenes. The dataset imposes several challenging factors including moving backgrounds, clutter scenes, scales and camera jitters. Experimental results show that our three-layer convolutional ISA network is able to learn features which are effective to represent complex activities such as human interactions in realistic environments.