This paper presents a new system architecture for managing multimedia content based on the digital item concept defined by MPEG-21. The focus is to provide a common interface such that: (1) media resources and metadata can be bound together and accessed seamlessly via a single platform-independent abstraction; (2) arbitrary, heterogeneous metadata formats can be used; and (3) content can be reliably exchanged and shared across different platforms. We demonstrate the effectiveness of such a system via two real-world applications: a media server product and a digital match management system for sports content.

This paper presents a new method for detecting semantic events which may happen in a soccer game. Our method requires the position information of the players and ball as input, which can be provided by a tracking system from processing either microwave signals or video sequences. Based on the tracking data, the method detects a wide range of player actions and game events using a set of heuristic rules which are derived from a hierarchical entity-relationship model representing the prior knowledge of soccer events. We have applied the proposed method as a key enabling module in a realworld sports entertainment application and the experimental results are given in the paper.

We propose a three-level video-event detection methodology and apply it to animal-hunt detection in wildlife documentaries. The first level extracts color, texture, and motion features, and detects shot boundaries and moving object blobs. The mid-level employs a neural network to determine the object class of the moving object blobs. This level also generates shot descriptors that combine features from the first level and inferences from the mid-level. The shot descriptors are then used by the domain-specific inference process at the third level to detect video segments that match the user defined event model. The proposed approach has been applied to the detection of hunts in wildlife documentaries. Our method can be applied to different events by adapting the classifier at the intermediate level and by specifying a new event model at the highest level. Event-based video indexing, summarization, and browsing are among the applications of the proposed approach.

We propose a three-level video event detection algorithm and apply it to animal hunt detection in wildlife documentaries. The first level extracts texture, color and motion features, and detects motion blobs. The mid-level employs a neural network to verify whether the motion blobs belong to objects of interest. This level also generates shot summaries in terms of intermediate-level descriptors which combine low-level features from the first level and contain results of mid-level, domain specific inferences made on the basis of shot features. The shot summaries are then used by a domain-specific inference process at the third level to detect the video segments that contain events of interest, e.g., hunts. Event based video indexing, summarization and browsing are among the applications of the proposed approach.

An algorithm is presented for replacing video background without a physical blue screen. The method requires pre-recording a background image of the scene without the presence of any foreground objects. Based on the color difference between the pixels in an input frame and their corresponding pixels in the background image, the algorithm computes a probability map which contains the likelihood for each pixel to be classified into the foreground or background. The probability map is further refined using anisotropic diffusion which reduces classification errors without introducing significant artifacts. Finally, based on the refined probability map, a new video is synthesized by feathering the foreground pixels of the input frames onto a new background video or image. The algorithm requires moderate computation resources and has been implemented to run in real time. Experimental results are presented.

This paper presents a statistics-based method for estimating the position and size of a face in a complex background. Face position and size are estimated based on robust statistical measurements which are derived from two one-dimensional histograms obtained by projecting the result of skin color filtering. The proposed algorithm also utilizes a linear Kalman filter and a simple nonlinear filter to perform smooth tracking and remove jitter. The algorithm has been implemented and tested under a wide range of real-world conditions. It has consistently provided performance which satisfies the following requirements: (1) the ability to automatically determine the initial position and size of a face and track it in a complex background; (2) is insensitive to partial occlusions and shadows; (3) is insensitive to face orientation and scale changes; and (4) is also insensitive to lighting condition changes. In addition, the algorithm is computationally simple so that it can be executed in real-time.

This paper presents a new scale, position and orientation invariant approach to object detection. The proposed method first chooses attention regions in an image based on the region detection result on the image. Within the attention regions, the method then detects targets using a novel object detection algorithm that combines template matching methods with feature-based methods via hierarchical MRF and MAP estimation. Hierarchical MRF and MAP estimation provide a flexible framework to incorporate various visual clues. The combination of template matching and feature detection helps to achieve robustness against complex backgrounds and partial occlusions in object detection. Experimental results are given in the paper.

In this paper, we present new methods for content-scalable shape representation and coding where different parts of a certain shape can be represented and coded at varying quality levels. We present a hierarchical vertex selection algorithm and develop a vertex-based content-scalable contour representation and coding method. In addition, we show how content-scalable shape coding can be achieved by modifying the leading bitmap-based shape coding methods. Experimental results are given in the paper.

This paper presents a new edge detection scheme that detects two-dimensional (2-D) edges by a curve-segment-based detection functional guided by the zero-crossing contours of the Laplacian-of-Gaussian (LOG) to approach the true edge locations. The detection functional is shown to be optimal in terms of signal-to-noise ratio (SNR) and edge localization accuracy; it also preserves the nice scaling property held uniquely by the LOG in scale space.

This paper presents a new two-dimensional (2D) edge detection scheme for general visual processing. The scheme constructs a 2D edge detection functional under the guidance of the Laplacian of Gaussian (LOG) zero crossing contours to detect edges. The detection functional is optimal in terms of signal-to-noise ratio (SNR) and edge localization accuracy (ELA) for detecting edges in 2D images; it also preserves the nice scaling property that is held uniquely by the LOG operator in scale space. The scheme also provides: (1) an edge regularization procedure; (2) an adaptive edge thresholding procedure; and (3) a scale space combining procedure. Experimental results on real images are given in the paper.

Studying the visual motion of human ambulatory patterns is of great importance for both vision research and applications. The authors first present three new algorithms for the recovery of three-dimensional structures from the motion of certain types of joints of articulated objects assuming perspective projection. Then based on the assumption that a human body can be considered as an articulated object and each part of the articulated object is rigid, they apply these algorithms to motion analysis of human ambulatory patterns in a combined way. The proposed algorithms have been tested in terms of the uniqueness of real solution and the speed of convergence using the synthesized data.>