This paper introduces a method for multidimensional Gaussian filtering using an efficient one-pass cascade of overlapping local-average windows driven by prefix sums. Each local-average filter is implemented in n dimensions, with non-integer lengths, allowing accurate approximation of Gaussians of any variance. In axis-oriented form the method has a scan-rate hardware realization and fast software implementation using minimal extra memory. In this latter case the new method consistently outperforms the fastest alternative Gaussian filtering method both in accuracy and speed.

The effectiveness of the cascaded object detector has been demonstrated repeatedly, and the basic steps to generate one are well known. But it is not obvious how to ensure the resulting detector will be efficient. This paper offers a solution by introducing a performance criterion that combines detection rate, false positive rate and speed, and a new cost function based on two simple, rational objectives. The result is a novel method for structuring the cascade which consistently produces efficient classifiers.

This paper describes a single-kernel conditional density estimation system that obtains descriptive parameters including gender, age, ethnicity, pose and expression from images of faces. The method is able to do fast estimation of 39 parameters from live video, achieving accuracies comparable with alternatives that yield only a few parameters. The single-kernel model can be interrogated to examine the properties of parameters and the training regime and thus guide design of more complicated estimators.

This paper presents a complete general-purpose method for still-image compression called adaptive prediction trees. Efficient lossy and lossless compression of photographs, graphics, textual, and mixed images is achieved by ordering the data in a multicomponent binary pyramid, applying an empirically optimized nonlinear predictor, exploiting structural redundancies between color components, then coding with hex-trees and adaptive runlength/Huffman coders. Color palettization and order statistics prefiltering are applied adaptively as appropriate. Over a diverse image test set, the method outperforms standard lossless and lossy alternatives. The competing lossy alternatives use block transforms and wavelets in well-studied configurations. A major result of this paper is that predictive coding is a viable and sometimes preferable alternative to these methods.

In appearance-based image processing, high-dimensional statistical models are estimated from low numbers of training samples. Sample scatter matrices are unreliable estimators of class covariances, yet many methods rely on them for dimensionality reduction and often for classification too. This paper argues for regularized covariance estimation and introduces a new method suitable for appearance-based image processing. The method is demonstrated for face detection, where a maximum likelihood classifier trained with regularized covariances achieves discrimination and detection results comparable to those of complicated multimodal and non-linear classifiers.

The expected value of missing data in a sample taken from a multivariate normal probability distribution is the mean of the conditional distribution of the missing dimensions given the known dimensions. We explain the derivation of this result, demonstrate its application to face image processing, then use it in a new method for recovering shape from image data. The context of our work is the use of 3D facial models to aid in recognition of human faces by humans. We explain the requirement for such models and review the practical possibilities for encoding depth information alongside photographs in identity documents like passports. The best alternative is to derive depths automatically from the photos, as this requires no side information. We show experimentally that conditional density estimation provides accurate face depth recovery, without recourse to explicit modelling of surface shape. 1

We describe an object-based video compression scheme based on the derivation and efficient coding of motion boundaries. First, we recursively identify a small number of global movement classes, each represented by two or more motion parameters. Second, we assign regions of a spatial segmentation to movement classes. Third, we merge segments using various similarity heuristics, while adding movement classes for small objects, if necessary. Finally, the boundaries of motion are coded using an efficient asymmetric binary tree coding scheme. Experimental results on standard test sequences qualitatively show that the proposed algorithm gives good segmentation, and that it is suitable for very-low-data-rate object-based coding.

There is wide application for the extraction of textual information from low-contrast, complex natural images. We are particularly interested in segmentation and thresholding algorithms for use in a portable text-to-speech system for the vision impaired. Reading low-contrast LCD displays is the target application. We present a low-complexity method for automatically extracting text of any size, font, and format from images acquired by a video camera that may be poorly focused and aimed, under conditions of inadequate and uneven illumination. The new method consists of fast thresholding that combines a local variance measure with a logical stroke-width method, and with a low-complexity statistical and contextual noise segmentation. The performance of this method compares favorably with more complex methods for the extraction of characters from scene images. Initial results are encouraging for application in a robust portable reader.

This paper describes a compression scheme for difference-image residues in video coding. Structured spatial patterns are used to map residue pixel values into a quadtree structure, which is then coded in significance order with the SPIHT algorithm. Thus the wavelet coefficient values of standard zerotree coding are replaced by untransformed (but carefully positioned) residue pixel values. The new zerotree pattern coding method compresses as well as zerotree wavelet coding and much better than DCT coding (as in MPEG) over error-free channels. Over noisy channels, zerotree pattern coding provides built-in error resilience, allowing transmission of residue data without error control overhead. A simple postprocessing technique provides additional error concealment.

Binary tree predictive coding (BTPC) is an efficient general-purpose still-image compression scheme, competitive with JPEG for natural image coding and with GIF for graphics. We report the extension of BTPC to video compression using motion estimation and compensation techniques which are simple, efficient, nonlinear and predictive. The new methods, binary tree recursive motion estimation coding (BTRMEC), and binary tree residue coding (BTRC) exploit the hierarchical structure of BTPC, in the first case giving progressively refined motion estimates for increasing numbers of pels and in the second case providing efficient residue coding. Compression results for BTRMEC and BTBC are compared against conventional block-based motion compensated coding as provided by MPEG. They show that both BTRMEC and BTRC are efficient methods to code video sequences.

Article Live paper: video augmentation to simulate interactive paper Share on Authors: Charles Robertson Multimedia Communications Lab, Faculty of Engineering & Applied Science, Memorial University of Newfoundland, St. John's, NF, Canada A1B 3X5 Multimedia Communications Lab, Faculty of Engineering & Applied Science, Memorial University of Newfoundland, St. John's, NF, Canada A1B 3X5View Profile , John Robinson Multimedia Communications Lab, Faculty of Engineering & Applied Science, Memorial University of Newfoundland, St. John's, NF, Canada A1B 3X5 Multimedia Communications Lab, Faculty of Engineering & Applied Science, Memorial University of Newfoundland, St. John's, NF, Canada A1B 3X5View Profile Authors Info & Claims MULTIMEDIA '99: Proceedings of the seventh ACM international conference on Multimedia (Part 2)October 1999 Pages 167–170https://doi.org/10.1145/319878.319923Online:01 October 1999Publication History 6citation526DownloadsMetricsTotal Citations6Total Downloads526Last 12 Months7Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

The keystroke dynamics of a computer user's login string provide a characteristic pattern that can be used for identity verification. Timing vectors for several hundred login attempts were collected for ten "valid" users and ten "forgers", and classification analysis was applied to discriminate between them. Three different classifiers were applied, and in each case the key hold times were more effective features for discrimination than the interkey times. Best performance was achieved by an inductive learning classifier using both interkey and hold times. A high rate of typographical errors during login entry is reported. In practice, these are usually corrected errors-that is, they are strings which include backspaces to correct earlier errors-but their presence confounds the use of typing-style analysis as a practical means of securing access to computer systems.

Binary tree predictive coding uses a noncausal, shape-adaptive predictor to decompose an image into a binary tree of prediction errors and zero blocks. Fast compression performance is comparable with Joint Photographers Expert Group (JPEG) for photographs, with Graphics Interchange Format (GIF) for graphics, and superior to the state of the art for composite images.

Virtual environments are often unsuitable for delicate operations because of the poverty of feedback, particularly tactile and force feedback.However, force feedback done properly requires large, heavy, expensive equipment.We have experimented with a particular form of tactile feedback using vibration as a substitute for force feedback.Substituting vibration for force feedback is intuitively appealing because it is cheap and low cost.But is it effective?Unfortunately, the answer is both "yes" and "no".We describe an experiment evahtating the effect of vibrotactile sensory substitution on user performance during a grasping task with delicate virtual objects.We found that adding vibrotactile feedback to visual and audio feedback improved task completion time for novice users, but led to increased grasp pressure over repeated uses of the system.

We report experiments on the 'foveation' of MPEGstandard video by a network-based rate adaptor.The rat e adaptor is controlled by input from receiving users, and constrained by the capacity of each receiver's channel.It generates versions of the input video which trade off spatial and temporal resolution in different regions of the picture.While there are many possible ways to make the tradeoffs, users have only one or two dimensions of control (in addition to two dimensions of position for the fovea centre).We therefore propose constraints for mapping 2D user control input into a fovea/periphery bit budget, and provide design curves for spatial resolution control.

A high-compression image coding scheme is presented, based on thread-like "Ridge" and "Valley" primitives. The use of these primitives is motivated by their success in economically representing image structure. The original image is sampled along the primitives, using a fractal yardstick method to determine sample spacing. The primitives themselves are compressed using vector coding and chain coding. Reconstruction at the receiver is a scattered data interpolation problem, solved here using C/sup 0/ natural neighbor interpolation. Results are presented showing data rates between 0.1 and 0.4 b/pixel, the degradations are discussed, and prospects for improvement outlined.>

Reports progress in primitive-based image coding using nonorthogonal dyadic wavelets. A 3D isotropic wavelet is used to approximate the difference-of-Gaussians (D-o-G) operator. Convolution of the image with dilated versions of the wavelet produces three band-pass signals that approximate multiscale smoothed second derivatives. An additional convolution of the image with a Gaussian-shaped low-pass wavelet creates a fourth subband signal that preserves low-frequency information not described by the three band-pass signals. The authors show that the original image can be recovered from the watershed and watercourse lines of the three band-pass signals plus the lowpass subband signal. By thresholding the watershed/watercourse representation, subsampling the low-pass subband, and using edge post emphasis, the authors achieve data reduction with little loss of fidelity. Further compression of the watersheds and watercourses is achieved by chain coding their shapes and predictive coding their amplitudes prior to lossless arithmetic coding. Results are presented for grey-level test images at data rates between 0.1 and 0.3 b/pixel.

Natural-image coding schemes may be classified into four families: pelwise, blockwise, region-based and model-based. The work described falls into the region-based category, but has several unique features. Space-curve primitives rather than surface and contour primitives are located at places of perceptual significance. An adaptive approach to the coding of primitives combines lossy yardstick compression of amplitudes and arithmetic coding of grouped directions. Bivariate natural neighbor interpolation is used for image surface recovery.>

The multimedia interactive conferencing application (MICA), a personal-workstation application for multipoint visual teleconferencing, is described. MICA allows people at two or more locations to share visual material such as documents, photographs, and computer screens in a highly interactive way. It supports the distribution, storage, retrieval, and high-quality display of visuals, real-time interaction by pointing and annotation, and meeting services facilities. The context of multimedia teleconferencing and computer-supported cooperative work is established, relating earlier research to the design of MICA. The services MICA offers are outlined. The handling, compression, and display of multiple media, and the design of a suitable user interface for MICA are discussed.>

The multipoint communication layer (MCL) is a generic multipoint communication service designed to support highly interactive multimedia conferencing applications. It supports full-duplex multipoint communication between an arbitrary number of connected locations over a variety of networks. MCL design was driven by the demanding real-time multipoint communication requirements of MICA (multimedia interactive conferencing application). The authors establish the context for MCL by considering MICA's communication requirements and reviewing related work in the area of multipoint communication. They present the overall design of the MCL service and describe its communication primitives, protocol message structures, and implementation. Comparison is made with alternative strategies and the use of MCL services for other applications is examined.>

A study is reported of the extraction of two-level cartoons from moving grey-level images for visual communication in the range 4.8-19.2 kbits/s. It is postulated that perceptually significant features of the human face and hands, at which cartoon lines should be drawn in the image, occur wherever surfaces in object space are approximately tangential to the line of sight of the camera or viewer. The implications of this postulate are analyzed for smooth lambertian reflectors seen against a background of the same luminance factor; it is shown that the combined effects of surface gradient and illumination lead to the formation of luminance valleys at the significant features. An experimental comparison of several detectors is described which confirmed the advantage of valley-based detection for deriving economical but recognizable cartoons.