Machine learning models, despite achieving low error rate, continue to encounter trust issues, particularly in domains where a single incorrect prediction can be costly or disastrous. WisdomNet architecture facilitates to achieve a zero error rate if certain conditions are met by rejecting data instances it is uncertain about and delegating those cases to a human expert. However, WisdomNet still faces several challenges, such as a high rejection rate and determining the appropriate point to stop fine-training. In this paper, we propose a novel technique called Difficulty-Driven Fine Training (DDFT), which not only determines when to stop fine-training but also minimizes the rejection rate. This technique focuses on excluding difficult data from the validation set, and fine-training WisdomNet with misclassified samples until the new validation set achieves a zero error rate. We conducted experiments to identify factors contributing to increased rejection rates in certain datasets. Our experimental results show that our method reduces the rejection rate from 39.90% to 30.53%, 96.62% to 23.25%, 89.09% to 9.80%, 5.24% to 1.65%, 94.4% to 20.55% in drybeans(Sira-Dermasons), banana quality, FashionMNIST(Dress-Shirt), MNIST(2-7) and CIFAR10(Cats-Dogs) datasets, respectively, compared to an ideal WisdomNet.

Recent studies focus on evaluating LLMs as judges to see if they match human-level capabilities. LLMs are used as judges in methods like pairwise comparison, single answer grading, and reference-guided grading. While traditional methods provide overall statistics of their performance, they may miss specific task weaknesses. LevelEval introduces an automated pipeline using rank-based evaluation to analyze LLMs across different input quality levels. This paper extends LevelEval by integrating a qualitative diagnosis report that assesses LLM strengths and weaknesses based on document types. This report includes difficulty scores for tasks and uses topic modeling to identify where LLMs excel or struggle during evaluation. Results show that this qualitative approach gives more insight than traditional metrics like BLEU and ROUGE. For instance, LLMs performed well on some topics but struggled with others. Improving open-source LLMs like Mistral on specific topics could enhance their evaluation performance in tasks like summarization and question answering, potentially matching GPT-4. The study found that Mistral underperformed on land-related law summarization and difficult math problems in retail and asset management. It also highlighted the need for better bias detection and classification in all open-source models to match GPT-4.

Convolutional neural networks (CNNs) have shown success in detecting Diabetic Retinopathy (DR) from retinal images captured by high-quality fundus cameras. As an alternate solution, smartphone-based systems with limited field-of-view (FoV) are recently proposed for DR screening. This paper investigates where to focus on the retina when using such smartphone-based devices. After training a CNN on original fundus images from diverse datasets, we evaluate the trained model on various regions of the retina (the fovea, the optic disc, the center of fovea and optic disc, the center of lower fovea and optic disc, and center of upper fovea and optic disc) that could be most effective to determine DR. Our experiments show that retinal images from smartphone-based systems with a narrower FoV (40%) that covers around fovea provided very close performance to original images with 0.963 AUC (within 0.99 of the optimum).

We gain insight to the COVID-19 pandemic response by the various U.S. states through analysis of open source emergency declaration, mitigation, and response policy data. We propose ASNM + POD, a Partial Ordering Detection extension to the Adaptive Sorted Neighborhood Method to identify redundancies and implied temporal ordering requirements to understand how various U.S. states respond to COVID-19. We further strengthen the well-established ASNM entity matching method and address key limitations of its Longest Common Subsequence extension (ASNM + LCS) through detection of all temporal order requirements. Partial order requirements are determined probabilistically through empirical review of all records’ time-ordered event sequences. We demonstrate effectiveness against a COVID-19 U.S. state policy dataset comprised of daily time-series data pulled from February and October 2022, where attributes are partially and variably populated. ASNM + POD yielded an F1 of 0.995 and an MCC of 0.985, significantly outperforming both ASNM and ASNM + LCS with F1/MCC improvements of 22%/50% and 15%/37%, respectively. Finally, we highlight the limited consensus on policies enacted, the variability in timelines of policy activations/deactivations, and activity at and after the two-year mark.

A vast range of Natural Language Processing (NLP) systems that are in use today have direct impact on humans. While machine learning models are expected to automatically infer world knowledge from historical texts, we should also be cognizant not to let NLP applications consume undesired societal stereotype bias back. Many bias evaluation measures have been designed and experimented to check whether unwanted stereo-type bias is present in the model or not. Upon performing various experiments, we found out that the most popular bias measures do not always indicate bias accurately. In addition to these experimental findings, we also propose our novel Differential Cosine Bias measure with examples of unwanted stereotype biases as well as necessary categorical bias that is based on knowledge. Our experiments show that our bias measure is a potential indicator of bias in NLP models compared to the popular bias evaluation measures.

In social interactions, we remember faces and associate names that we hear to the faces. Existing research has studied single-item memory tasks such as names, face, colors, shapes, and classified their associations using spectral analysis, event-related potentials(ERPs), and non-ensemble machine learning techniques. However, the impact of multiple stimuli presentation modalities in the associative memory of face-name stimuli has not been extensively investigated. In this research, we conducted two experiments on audiovisual face-name and visuospatial face-name memory tasks. We recorded EEG data from 15 healthy participants and extracted the average oscillatory band power of the alpha, theta, beta, and gamma bands and the signal entropy. Adaptive Logistic boosting was applied for classification, and t-test statistical analyses were applied to investigate participants’ behavioral performance. We found that people did better on visuospatial face-name memory association than audiovisual face-name memory association. Adaptive boosting performed best in comparison to other existing algorithms with the average classification accuracy of 76.37% at the parietal electrode P7.

Electroencephalography(EEG) signals have been used to assess the efficiency of recall based on the evaluation of memories remembered or forgotten. There is need for an EEG noise/outlier resistant algorithm with a good classification accuracy. In this paper, we present a new technique for EEG signal classification to study whether a person remembers faces he or she has seen or not. For this purpose, our method applies denoising using discrete wavelet transform, time-locked EEG activity before generating templates for each class that indicates whether a person remembered or forgot. This novel technique was compared with well-known classification algorithms such as Discriminant Analysis (DA), Support Vector machines (SVM using linear kernel) and Decision Trees (DT). Our Template Matching for EEG Single trial (ETMEST) technique outperformed them all. For all subjects, the average classification accuracy of ETMEST was 94%. Also, we applied this technique on a visually evoked potential EEG dataset based on c-VEP BCI speller system to detect symbol/alphabet recognition events. The results show this method can be reliably applied to EEG-based Communication, Brain Computer Interface(BCI) applications and other signal classification problems. It is additionally surmised from this investigation that during memory encoding, the signals from the frontal and temporal lobes are the strongest and during decoding, signals from the frontal and parietal lobes are the strongest.

The crystallization of biological macromolecules like proteins is an important process to study their molecular structures. The quality of crystals is critical to be able to determine their structures using methods such as X-ray crystallography. Therefore, many wet-lab experiments are conducted using numerous screening plates to obtain successful crystal growth. High-throughput microscopy is useful to quickly collect images from the screening plates. Since the automated systems for imaging require high-end instrumentation, they are costly. This study investigates a small scale, mobile fluorescence imaging system, and application. Our system is composed of a mobile imaging system, a mobile app to capture images from plates, and a machine learning model to recognize the presence of crystals presence from images. For fluorescence imaging, we present an assembly of a smartphone or tablet integrated with a macro lens tube and illumination LEDs. The system presented in this study has magnification range from 20x to 250x macro. For the recognition of crystals, a convolutional neural network model was trained on a computer and then deployed on the mobile app. A data set of 1000 trace fluorescently labeled images was used to train and evaluate the model. The accuracy of the hold-out testing images was about 95%. The mobile app for imaging and protein recognition was developed to run on Apple iOS devices. To evaluate the system further, the recombinant inorganic pyrophosphatase protein from Klebsiella pneumoniae, which was expressed from E. coli, was crystallized using the trace fluorescent labeling method. Our system can capture quality images of protein crystals in both white and fluorescence lights. The overall accuracy of recognizing crystal or non-crystal outcomes on the pilot test is about 93%. This mobile imaging system can be useful for small group research labs and students.

The data representation as well as naming conventions used in commercial screen files by different companies make the automated analysis of crystallization experiments difficult and time-consuming. In order to reduce the human effort required to deal with this problem, we present an approach for computationally matching elements of two schemas using linguistic schema matching methods and then transform the input screen format to another format with naming defined by the user. This approach is tested on a number of commercial screens from different companies and the results of the experiments showed an overall accuracy of 97 percent on schema matching which is significantly better than the other two matchers we tested. Our tool enables mapping a screen file in one format to another format preferred by the expert using their preferred chemical names.

Protein crystallization well plate is a rectangular platform that contains wells usually organized as a grid structure. The crystallization conditions are studied through a screening process by setting up the trial conditions in the well plate. In the past, the expert evaluates the trial wells for the growth of crystals by manually viewing the plate under a microscope or using a high-throughput plate imaging and analysis system. While the first method is tedious and cumbersome, the second method requires financial investment. Recently, a few approaches were developed by collecting images using smartphones thus enabling low-cost automatic scoring (classification) of well images. Nevertheless, these recent methods do not detect which well on the plate is captured. If the user has a smartphone, the user may capture or scan any well by just moving the smartphone to the corresponding well. In this paper, we propose a mobile scanner that identifies the well by using a coded template under the well plate. The mobile scanner provides two modes: image and video. Image mode is used for single well analysis whereas video mode is used to scan the complete plate. In the video mode, the mobile scanner app generates a tilemap of the plate.

Recently, mirage pose estimation method was proposed for multi‐camera systems. Multi‐camera mirage analytically solves a system of linear equations for six pose parameters in O( n ) time. Mirage promises to execute in real time with high accuracy and shows lower rotational and translational errors compared to eight other well‐known perspective‐n‐points (PnP) methods. However, the simulated tests and real experiments showed that, in case of a single camera, the analytical system of linear equations is not solvable due to the reduced rank of the linear system that is obtained by the formulation. In this study, an important revision to mirage is proposed to support single camera systems properly. The results of simulations and real experiments demonstrate smaller pose estimation errors compared to a group of eight well‐known state‐of‐the‐art PnP methods.

Tropical cyclone intensity estimation is a challenging task as it required domain knowledge while extracting features, significant pre-processing, various sets of parameters obtained from satellites, and human intervention for analysis. The inconsistency of results, significant pre-processing of data, complexity of the problem domain, and problems on generalizability are some of the issues related to intensity estimation. In this study, we design a deep convolutional neural network architecture for categorizing hurricanes based on intensity using graphics processing unit. Our model has achieved better accuracy and lower root-mean-square error by just using satellite images than 'state-of-the-art' techniques. Visualizations of learned features at various layers and their deconvolutions are also presented for understanding the learning process.

Protein crystallization is the process of formation of protein crystals. Many combinations of chemicals need to be tried to obtain a crystal for some difficult proteins. This paper discusses a novel way of identifying the various conditions necessary for a successful crystal growth by using a variation of genetic algorithm which explores unexplored territories of the chemical search space, thereby increasing the probability of finding a new crystalline condition. Our analysis shows that our method has 6 common crystalline conditions with associative experimental design(AED) and 9 common crystalline conditions with GenScreen for the protein AbIPPase.

The presence of heterogeneity in the data representation of the commercial protein screen makes the analysis of experimental results and the protein crystallization screening process difficult and time-consuming. In order to reduce the human effort required to deal with this problem, we present our application based on schema matching and data integration to automatically find the matching elements between the two screen schemas and then transform the input screen file to the required output screen format.

Protein crystallization experiments require the inspection and analysis of a large number of crystal images. So, manual inspection of these images is very inefficient and might risk missing successful results. Visual-X2 is a tool developed to aid the user for quick and efficient visualization and analysis of the results of the experiments. In this paper, we discuss various features of Visual-X2 and their applicability in the crystallization experiment.

This paper presents a new method of segmenting and classifying protein crystallization trial images that were collected using trace fluorescent labeling. Trace fluorescent labeling typically involves fluorescence dye that can re-emit the illumination light at other wavelengths around the principal wavelength. The captured image has a primary color channel with respect to illumination light and fluorescence dye. Crystals will have higher intensity than non-crystal areas. But there might be bright regions that may not be crystals, thereby making inaccurate and not robust trial images classification. In this paper, we utilize the subordinate color channel besides the primary color in the image of trace fluorescently labeled protein solution. This new method extracts proper features and successfully builds a high accuracy classifier with a low rate of misclassification of crystals as non-crystals. We also present a framework that could optimize both image segmentation and classification. In our experiments, we achieved around 94% accuracy with 0.6% misclassification of crystals as non-crystal.

Artificial Neural Networks are a widely used computing system implemented for a wide variety of tasks and problems. A common application of such networks is classification problems. However, a significant amount of this research focuses on one and two-dimensional information, such as vectorized data and images. There is limited research performed on three-dimensional media such as video clips. This can be attributed to a lack of adequate resources, available training datasets, hardware constraints, and appropriate frameworks for implementing such networks. This paper attempts to provide an alternate methodology of feeding three-dimensional video data by preprocessing instead of directly inputting to a deep convolutional neural network. By taking sequential segments from multiple frames of a single video clip and combining them into a single image, the temporal dimension of the video can be encoded as a two-dimensional image. This process is called as temporal slicing and repeated for the entire spatial dimension of the video. The end result is spatio-temporal data encoded in a spatial format, which is then propagated through a convolutional neural network as image data. This method is less resource-intensive and is remarkably faster than pre-existing three-dimensional convolutional methods, while achieving significantly higher accuracy compared to the aforementioned network architectures.

In general, a single thresholding technique is developed or enhanced to separate foreground objects from background for a domain of images. This idea may not generate satisfactory results for all images in a dataset, since different images may require different types of thresholding methods for proper binarization or segmentation. To overcome this limitation, in this study, we propose a novel approach called "super-thresholding" that utilizes a supervised classifier to decide an appropriate thresholding method for a specific image. This method provides a generic framework that allows selection of the best thresholding method among different thresholding techniques that are beneficial for the problem domain. A classifier model is built using features extracted priori from the original image only or posteriori by analyzing the outputs of thresholding methods and the original image. This model is applied to identify the thresholding method for new images of the domain. We performed our method on protein crystallization images, and then we compared our results with six thresholding techniques. Numerical results are provided using four different correctness measurements. Super-thresholding outperforms the best single thresholding method around 10 percent, and it gives the best performance for protein crystallization dataset in our experiments.

Unmanned vehicles are autonomous robotic systems that are fully or partially controlled by an operator remotely from a station. In the last two decades, massive amount of advancements have been observed regarding unmanned vehicles for both military and civilian purposes. Today majority of these vehicles require human guidance even for basic missions, thus, minimising the human intervention on such systems is one of the emerging research topics. To serve this purpose, this study proposes a new trajectory tracking algorithm using Mirage pose estimation method. Mirage employs target pixel errors in two‐dimensional image plane and analytically calculates the robot's pose in three‐dimensional Euclidean space. Therefore, complex computations are not needed and undesirable Euclidean trajectories are avoided since the vehicle's pose is directly controlled. Both simulations and real experiments were performed to verify the effectiveness of the method. The results show that the proposed method is a feasible alternative for vision‐based Euclidean trajectory tracking with high accuracy and low complexity.

Automated image analysis of microscopic images such as protein crystallization images and cellular images is one of the important research areas. If objects in a scene appear at different depths with respect to the camera's focal point, objects outside the depth of field usually appear blurred. Therefore, scientists capture a collection of images with different depths of field. Focal stacking is a technique of creating a single focused image from a stack of images collected with different depths of field. In this paper, we introduce a novel focal stacking technique, FocusALL, which is based on our modified Harris Corner Response Measure. We also propose enhanced FocusALL for application on images collected under high resolution and varying illumination. FocusALL resolves problems related to the assumption that focus regions have high contrast and high intensity. Especially, FocusALL generates sharper boundaries around protein crystal regions and good in focus images for high resolution images in reasonable time. FocusALL outperforms other methods on protein crystallization images and performs comparably well on other datasets such as retinal epithelial images and simulated datasets.

This paper presents a GPU based image registration algorithm that utilizes Hough Transform and Least Square Optimization to calculate the transformation between two images. In our approach, we calculate the transformation parameters of all possible combination solutions of matched feature points by exploiting parallel processing power of the GPU. We applied our algorithm on a variety of images including the problem of mosaic image generation. Experimental results show that our method is robust to the outliers (incorrect matches) and it can achieve very accurate registration (numeric and visual) results with much faster (up to 20 times) than CPU implementation.

Multi-class classification problem has become a challenging problem in bioinformatics research. The problem becomes more difficult as the number of classes increases. Decomposing the problem into a set of binary problems can be a good solution in some cases. One of the popular approaches is to build a hierarchical tree structure where a binary classifier is used at each node of the tree. This paper proposes a new greedy technique for building a hierarchical binary classifier to solve multiclass problem. We use neural networks to build all possible binary classifiers and use this greedy strategy to build the hierarchical tree. This technique is evaluated and compared with two popular standard approaches One-Versus-All, One-Versus-One and a multi-class single neural network based classifier. In addition, these techniques are compared with an exhaustive approach that utilizes all possible binary classifiers to analyze how close those classifiers perform to the exhaustive method.

There has been related research for sprite or mosaic generation for over 15 years. In this article, we try to understand the methodologies for sprite generation and identify what has not actually been covered for sprite generation. We first identify issues and focus on the domain of videos for sprite generation. We introduce a novel sprite fusion method that blends two sprites. Sprite fusion method produces good results for tracking videos and does not require object segmentation. We present sample results of our experiments.

Global motion estimation (GME) is a critical step for image alignment, image registration, and sprite generation. Direct methods use all pixels to estimate the motion. Eliminating pixels for GME is important since it may reduce the processing time and may also help to obtain correct motion parameters. In this paper, we firstly consider using fixed masks to observe the performance of GME. Then, we generate and use texture masks to eliminate texture regions to improve the performance of GME. The texture regions may include water, grass, ground, sky, etc. Our results indicate that adapting suitable masks reduces the processing time and improves the correctness of GME.

Without the ground truth image, there is no proper objective method to evaluate sprite generation. In this paper, we propose several complex camera motion patterns to generate synthetic video from original images. Our camera motion patterns include Rotation, Affine, and Pan-Tilt-Zoom (PTZ) transformations. In addition, they also include combined patterns. Subsequently, we applied sprite generation to the synthetic videos. Objective evaluation is performed by comparing the ground truth image and sprite based on Peak-Signal-to-Noise Ratio (PSNR) and size. Pattern algebra file is also provided for estimating the accuracy of sprite generation with respect to global motion parameters. Our result indicates that frame PSNR, picture PSNR, size, ground truth image, and pattern algebra file are good indicators of sprite quality.

SpriteDB tool has two main objectives. Firstly, it provides an open-source, uniform platform for accessing a vast collection of test videos used for sprite generation. Secondly, it presents the experiments with intermediate results generated by our approach for comparison and analysis. Our experimental results consist of classification of videos for sprite generation, synthetic and ground truth video generation, and the generated sprites.

Abstract Correctness of concurrent software is usually checked by techniques such as peer code reviews or code walkthroughs and testing. These techniques, however, are subject to human error, and thus do not achieve an in‐depth verification of correctness. Model‐checking techniques, which can systematically identify and verify every state that a system can enter, are a powerful alternative method for verifying concurrent systems. However, the usefulness of model checking is limited because the number of states for concurrent models grows exponentially with the number of processes in the system. This is often referred to as the ‘state explosion problem.’ Some processes are a central part of the software operation and must be included in the model. However, we have found that some exponential complexity results due to uncontrolled concurrency introduced by the programmer rather than due to the intrinsic characteristics of the software being modeled. We have performed tests on multimedia synchronization to show the effect of abstraction as well as uncontrolled concurrency using the Promela/SPIN model checker. We begin with a sequential model not expected to have exponential complexity but that results in exponential complexity. In this paper, we provide alternative designs and explain how uncontrolled concurrency can be removed from the code. Copyright © 2007 John Wiley & Sons, Ltd.

Abstract With the increasing need for different energy saving mechanisms in Wireless Sensor Networks (WSNs), data aggregation techniques for reducing the number of data transmissions by eliminating redundant information have been studied as a significant research problem. These studies have shown that data aggregation in WSNs may produce various trade‐offs among some network related performance metrics such as energy, latency, accuracy, fault‐tolerance and security. In this paper, we investigate the impact of data aggregation on these networking metrics by surveying the existing data aggregation protocols in WSNs. Our aim is twofold: First, providing a comprehensive summary and comparison of the existing data aggregation techniques with respect to different networking metrics. Second, pointing out both the possible future research issues and the need for collaboration between data management and networking research communities working on data aggregation in WSNs. Copyright © 2006 John Wiley & Sons, Ltd.

Multimedia presentations are the basic objects of multimedia databases. Since a multimedia presentation is not an instant display of a query result, the control knowledge (or synchronization requirements) has to be incorporated into the database and necessary precautions have to be taken for a lengthy presentation. Active databases provide a mechanism for incorporation of control knowledge by using event-condition-action (ECA) rules. In this paper, we describe how multimedia synchronization can be handled within a database using ECA rules. We present a prototype presentation synchronization database, named as PressBase, for distributed multimedia systems. We have adopted one of the synchronization models, SynchRuler, and then incorporated into a relational database system.

Peer-to-peer (P2P) streaming supports distributed data transfer over Internet. It claims high resource utilization and better streaming performance. In this paper, we define mesh-cast concept to model "many-to-many" streaming in P2P networks. The main origin of network congestion in mesh-cast is the articulation point/edge in inferred network topology. We resolve the congestion using the peripheral articulation node. We provide two possible approaches, optimistic and pessimistic, to solve network congestion in mesh-cast, both of which use a novel P2P structure, G-Super-Peer backbone.

Flexible synchronization models cannot provide a proper way of managing user interactions that change the course of a presentation. In this paper, we present a flexible synchronization model, termed SynchRuler, which allows such user interactions including backward and skip. The synchronization rules, which are based on event-condition-action (ECA) rules, are maintained to handle relationships among streams in SynchRuler. The synchronization rules are manipulated by the receiver-controller-actor (RCA) scheme, where receivers, controllers, and actors are objects to receive events, to check conditions, and to execute actions, respectively. The verification of a multimedia presentation specification is performed with the synchronization model. The correctness of the model and the presentation is controlled with a technique called model checking. Model checker PROMELA/SPIN tool is used for automatic verification of the correctness of LTL (linear temporal logic) formulas.

The ordinary sprite generation techniques focus on camera movement, accurate motion estimation, alignment, and integration. These techniques ignore the resolution of the original images, and the regenerated images from the sprite are likely to have lower resolutions than the original ones. Especially if the scenes have finite depth and zoom-in and zoom-out operations occur, the segments of the scene are captured at different resolutions. The traditional mosaic generation methods either blur the mosaic by integrating lower resolution segments or use unnecessarily large storage for the mosaic. The sprite pyramid (or layered sprite) allows efficient storage of images or video clips of overlapping scenes at different resolutions. Moreover, the images or video frames can be reconstructed from the sprite pyramid at the necessary resolutions

Virtual camera controls (VCC) for digital video enable the viewers to visualize interesting objects from their perspective. VCC also allow us to play the video from different angles. In this sense, VCC do not only support visualization and browsing capabilities but also support playback from different angles. VCC management requires accurate global motion estimation and accurate sprite generation. After motion parameters are detected and the precise sprite is generated, virtual camera controls are used to manipulate the sprite and the original frames to allow interactive spatial browsing and playback that enables panning, tilting, and zooming.

Global motion estimation (GME) techniques have been developed and usually applied on video where motion takes place often. Although these methods produce accurate results where frequent motion occurs, they turn out to be inefficient if motion is not so often in the video, as in semi-dynamic video. In this paper, we propose motion sensors that indicate the existence of motion and yield a quick approximation to the motion when motion exists thus removing the computations of the hierarchical evaluation of low-pass filtered images as in iterative descent methods.

We present a method to extract coarse boundary features from discrete cosine transform (DCT) compressed blocks. Video sequences usually contain enormous data for video processing. Decompressing data and then processing uncompressed data is computational intensive. Moreover, this may yield erroneous results due to processing unnecessary data. Our goal is to extract coarse boundary features without decompressing original data, and then to eliminate insignificant blocks for decompressing. The extracted features provide information about the smoothness, the boundary visibility, and the boundary structure of a block. We give an application of using coarse features for video object segmentation.

The mosaic generation methods benefit from global motion estimation (GME) methods, which yield almost accurate estimation of motion parameters. However, the generated mosaics are usually more blurred than the original frames due to the image warping stage and errors in motion estimation. The transformed coordinates resulting from GME are generally real numbers whereas images are sampled into integer values. Although GME methods generate proper motion parameters, a slight error in motion estimation may propagate to subsequent mosaic generation steps. We propose a method to generate clearer mosaics from video. The temporal integration of images is performed using the histemporal filter based on the histogram of values within an interval. The initial frame in the video sequence is registered at a higher resolution to generate a high resolution mosaic. Instead of warping of each frame, the frames are warped into the mosaic at intervals. This reduces the blurring in the mosaic.

We present a new flow and congestion control scheme, PLUS (Probe-Loss Utilization Streaming protocol), for distributed multimedia presentation systems. This scheme utilizes probing of the network situation and an effective adjustment mechanism to data loss to support multimedia presentations. The proposed scheme is also designed to scale with increasing number of PLUS-based streaming traffic and to live in harmony with TCP-based traffic. The novelty of the PLUS protocol is that it utilizes the knowledge of its future bottleneck bandwidth in probing the current network situation. This can be achieved by a priori knowledge of the multimedia data before a presentation is requested by a client. Compression schemes like MPEG introduce dependencies on media units. I frames are needed to successfully decode P and B frames, and P frames are needed to decode B frames. A loss of an I or P frame automatically eliminates dependent media units. Our probing scheme increases the successful transmission of critical I and P packets without the overhead of error-correction-schemes. Probing is done using B-frame packets. The advantage is that we use data packets as probe packets. With the PLUS protocol we address the need to avoid congestion rather than react to it. Experiments demonstrate the effectiveness of the approach in utilizing network resources and decreasing loss ratios.

The development of the new video coding standard, MPEG- 4, has triggered many video segmentation algorithms that address the generation of the video object planes (VOPs). The background of a video scene is one kind of VOPs where all other video objects are layered on. In this paper, we propose a method for the generation of the stationary background in a MPEG compressed video sequence. If the objects move frequently and all the components of the background are visible in the video sequence, the background macroblocks can be constructed by using Discrete Cosine Transform (DCT) DC coefficients of the blocks. After the generation of the stationary background, the moving objects can be extracted by taking the difference between the frames and the background.

An interactive multimedia presentation in a distributed multimedia system requires synchronization of media streams, preprocessing media for content-based retrieval and low-bandwidth transmission over network, and user interface for interacting multimedia presentations. The power of synchronization models is limited to the synchronization specifications and user interactions. We propose an event-based synchronization model that can handle time-based actions while enabling user interactions like backward and skip. For effective transmission of multimedia data, the multimedia data needs to be preprocessed. The sprite generation and moving objects segmentation can reduce the required bandwidth significantly. We propose a method for multiresolution sprite which will allow reproduction of the video at different resolutions. The object segmentation will be extracted by generating a closed boundary for the object. Since the video data may also exist in a compressed format, we also propose to extract new features from the compressed video. We will consider compressed data that is generated by Discrete Cosine Transform (DCT) which has been used in MPEG-1, MPEG-2, MPEG-4 \cite{MPEG-4} and H263.1. The user will be provided a high-level user interface to access the contents of the presentation. We will test this integrated framework on distance education project over the Internet.

The gap between the synchronization specification and the synchronization model limits user interactions for a multimedia presentation. The middle-tier for multimedia synchronization handles the synchronization rules that are directly extracted from the specification. In addition to these rules, the middle-tier also manages implicit synchronization rules which are not specified but can be extracted from other rules. The synchronization rules generated by the middle-tier assists the synchronization model to provide user interactions while keeping the synchronization specification minimal. We give examples of how these rules are generated from SMIL expressions.

Advances in multimedia computing offer new approaches to support online access to information from a variety of sources such as video, audio, images, and transparencies. The key to success in providing distributed real time retrieval, is the control of network delay and congestion to keep up the desired quality of service (QoS). We present a new distributed multimedia database environment, called NetMedia. This environment can support controlled delivery of video, audio, and text data, across Internet. NetMedia can be the base of various distributed multimedia applications. As an example we present NetMedia-Virtual-Classroom, an asynchronous distance learning tool, which uses the advantages of the NetMedia framework.