Catálogo de publicaciones - libros

The inherent noises can significantly degrade the accuracy of the attention area detection. In this paper, we present a novel structure of hybrid filter for suppressing noises in saliency maps which is viewed as a preliminary step towards the solution of automatic video region-of-interest determination. The filter presented in our paper makes use of the property of saliency maps and can remove almost all the Gauss noise and pepper-salt noise while preserve the details of attention area. Experimental results demonstrate the efficiency and effectiveness of our approach in extracting the region of interest.

The various techniques used to determine the reduced number of features in principal component analysis are usually ad-hoc and subjective. In this paper, we use a method of finding the number of features which is based on the saturation behavior of a graph and hence is not ad-hoc. It gives a lower bound on the number of features to be selected. We use a database of handwritten digits and reduce the dimensions of the images in this database based on the above method. A comparison with some conventional methods such as scree and cumulative percentage is also performed. These two methods are based on the values of the eigenvalues of the database covariance matrix. The Mahalanobis and Bhattacharyya distances will be shown to be of little use in determining the number of reduced dimensions.

A new video object encoding algorithm based on the curve fitting trajectory of video object moving edges pixels is proposed in this paper. This algorithm exploits the fact that, under certain circumstances where the objects in the video are not moving quickly as in video conferencing and surveillance, the only significant video object information are the edges of the moving objects. Other object information remains relatively constant over time. This algorithm is modified from the standard video compression algorithms. Simulation results show that, under standard test sequences, this encoding algorithm has a lower bit rate than the classical DCT method.

Under the limited bandwidth, MPEG-4 video coding stream with Fine Granularity Scalability can be flexibly dropped by very fine granularity to adapt to the available network bandwidth. Therefore, we can either reduce the frame rate, i.e., reduce the frames per second(FPS), by dropping partial frames to keep the spatial sharpness of an image or reduce the bits per frame(BPF) to keep the temporal continuity of a video. We attempt to understand that different content characteristics for the above two schemes affect the visual perceived quality when the bandwidth is limited. In this paper, the double stimulus continuous quality evaluation(DSCQE) is used as our subjective measurement. In our experiment, the subjects assess the scores of perceived quality by comparing the reference sequences with the test sequences for different content characteristics. We find that video contents with low motion characteristic suit to low frame rate and video contents with high motion characteristic suit to high frame rate under the limited bandwidth. The perceived quality of the spatial sharpness for the detailed texture sequences is influenced more than the easy texture sequences when the bit rate is increased.

The gray levels quantization technique is used to generate images which limit the number of color levels resulting in a reduction of the image size, while it preserves the quality perceived by human observers. The problem is very relevant for image storage and web distribution, as well as in the case of devices with limited bandwidth, storage and/or computational capabilities. An efficient evolutionary algorithm for the inverse gray level quantization problem, based on a technique of dynamical local fitness evaluation, is presented. A population of blur operators is evolved with a fitness given by the energy function to be minimized. In order to avoid the unfeasible computational overhead due to the fitness evaluation calculated on the entire image, an innovative technique of dynamical local fitness evaluation has been designed and integrated in the evolutionary scheme. The sub–image evaluation area is dynamically changed during evolution of the population, and the evolutionary scheme operates a form of machine learning while exploring subarea which are significatively representative of the global image. The experimental results confirm the adequacy of such a method.

In this paper, a novel framework is proposed to enable intuitive mining and exploration of large-scale video news databases via knowledge visualization. Our framework focuses on two difficult problems: (1) how to extract the most useful knowledge from the large amount of common, uninteresting knowledge of large-scale video news databases, and (2) how to present the knowledge to the users intuitively. To resolve the two problems, the interactive database exploration procedure is modeled at first. Then, optimal visualization scheme and knowledge extraction algorithm are derived from the model. To support the knowledge extraction and visualization, a statistical video analysis algorithm is proposed to extract the semantics from the video reports.

The patterns of missing values are important for assessing the quality of a classification data set and the validation of classification results. The paper discusses the critical patterns of missing values in a classification data set: missing at random, uneven symmetric missing, and uneven asymmetric missing. It proposes a self-organizing maps (SOM) based cluster analysis method to visualize the patterns of missing values in classification data.

This paper presents a keyword-based information visualization technique for unstructured text sequences. The text sequence data comes from nursing narratives records, which are mostly text fragments with incomplete and unreliable grammatical structures. Proper visualization of such text sequences can reveal patterns and trend information rooted in the text records, and has significant applications in many fields such as medical informatics and text mining. In this paper, an Iterative Visual Clustering (IVC) technique is developed to facilitate multi-scale visualization, and at the same time provide abstraction and knowledge discovery functionalities at the visualization level. Interactive visualization and user feedbacks are used to iteratively group keywords to form higher level concepts and keyword clusters, which are then feedback to the visualization process for evaluation and pattern discovery. Distribution curves of keywords and their clusters are visualized at various scales under Gaussian smoothing to search for meaningful patterns and concepts.

The goal of clustering in data mining is to distinguish objects into partitions/clusters based on given criteria. Visualization methods and techniques may provide users an intuitively appealing interpretation of cluster structures. Having good visually separated groups of the studied data is beneficial for detecting cluster information as well as refining the membership formation of clusters. In this paper, we propose a novel visual approach called , based on the projection technique of HOV to achieve the separation of cluster structures. With M-mapping, users can explore visual cluster clues intuitively and validate clusters effectively by matching the geometrical distributions of clustered and non-clustered subsets produced in HOV.

While the searching of text document has grown relatively mature on the Internet, the searching of images and other forms of multimedia data significantly lags behind. To search visual information on the basis of semantic concepts requires both their discovery and meaningful indexing. By analyzing the users’ search, relevance feedback and selection patterns, we propose a method which allows semantic concepts to be discovered and migrated through an index hierarchy. Our method also includes a robust scoring mechanism that permits faulty indexing to be rectified over time. These include: (i) repeated and sustained corroboration of specific index terms before installation, and (ii) the ability for the index score to be both incremented and decremented. Experimental results indicate that convergence to an optimum index level may be achieved in reasonable time periods through such dynamic index evolution.