The Digital Media Laboratory has spent a significant portion of 2009 exploring the creation of theatrical performances in the online world Second Life. This has taught us practical methods for direction and stagecraft in virtual spaces, and has shown the advantages and limitations of such performances. A virtual space has the advantages of being easily accessible, flexible, and tolerant of mistakes - an scenery can be taken down and rebuilt in minutes. Set design can be done in three dimensions, and the first draft is often a prototype of what will actually be used. Some scripts that would otherwise be impossible to mount in a conventional theatre are relatively easily produced in Second Life. This kind of performance in a virtual arena raises provocative questions regarding future directions in theatre design, delivery, education and the very nature of performance. Is Second Life the likely arena for a new wave of alternative theatre?

Hand pose and gesture recognition methods have been based on many properties of hand images, but using finger recognition is unique. Here we describe hand pose from finger recognition, its limitations and its advantages. The implementation is being used for some real applications, including interfaces and computer games, and has a success rate of over 98%.

Gaze tracking is the complex process of detecting where a person is looking. In this paper, we present a gaze tracking system that estimates the eye gaze by using a stable reference point created through the extraction of the sclera - otherwise known as the ¿whites¿ of the eye.

A combination of three algorithms is proposed that gives a reasonable rate of success in image similarity searches. All use region based statistical measures using a combination of color features, linear features, and intensities combined with a collection of methods for subdividing the image into regions. All possible combinations were evaluated and the best (highest success rate) was selected to be used as a practical image similarity measure. Component algorithms are combined using a simple decision fusion voting scheme, giving success rates in the range of 50–60%. For a web search application this is quite reasonable.

When creating multimedia constructions such as Web pages, games, and animations, it is common to play some sorts of sound as a 'loop', where the sound file is played repeatedly from beginning to end. This is an easy way to extend a short audio sample, but the repetition is easily detected by a human listener. It is possible to create new sounds from examples, so that looping is not needed.

Sending audio data over a computer network consumes a large amount of bandwidth, and so compression strategies are regularly built into audio file formats and transmission software. In some environments, the basic nature of the sound does not change significantly; for example, phone lines deal frequently with voice transmission. By matching input audio blocks against those in a table, we can transmit the table indices only, and audio can be synthesized at the receiving end by simple table look-up. This has a number of potentially interesting applications.

Two approaches to setting programming assignments are compared, using actual assignment and solutions from two distinct CS102 lecture sections during the same semester. The complexity and effort represented by the solutions is compared using software engineering metrics, giving a measure of the effectiveness of the two assignments sets, and student performance in successor courses is also examined.

We explain the need for formal collaboration between students in the arts and computer science students, and argue that this should be encouraged at an early stage of their education. Two specific courses that foster this sort of cooperation are described.

No abstract available.

This contest involved the running and evaluation of computer vision and pattern recognition techniques on different data sets with known groundwidth. The contest included three areas; binary shape recognition, symbol recognition and image flow estimation. A package was made available for each area. Each package contained either real images with manual groundtruth or programs to generate data sets of ideal as well as noisy images with known groundtruth. They also contained programs to evaluate the results of an algorithm according to the given groundtruth. These evaluation criteria included the generation of confusion matrices, computation of the misdetection and false alarm rates and other performance measures suitable for the problems. The paper summarizes the data generation for each area and experimental results for a total of six participating algorithms.

Template matching is rarely used for symbol recognition in practice because of the variation in scale, orientation, and line thickness and position found in actual images. An answer to this is to use an adaptable template, one that can be scaled and rotated easily. Here the use of such a template, based on vector representation of symbols, is explored. Results from handprinted digits and electronic symbols are presented.

A method for reconstructing 3D models from cross-sectional signals is presented. This is specifically useful for low-resolution images (e.g. seismic or ultrasound images). The method is not voxel-based, and is faster and uses less memory than methods that are. The algorithm starts by pre-processing the images by thresholding, segmenting, contour finding, point sampling and then triangulating. The main objective of this algorithm is to produce the smoothest possible model for branching bodies, taking into consideration the fact that the slices might be far apart in reality, and consequently a novel branching/interpolation technique is used that proved superior to existing methods for this purpose. At the same time, it did not need the involvement of the user for objects within defined classes. Seismic simulations and CT data have been used to test the algorithm. We succeeded in reconstructing the tomb of the Egyptian Pharaoh Tutankhamen from simulated seismic data.

This paper provides an overview to the implementation of Lemon, a complete optical music recognition system. Among the techniques employed by the implementation are: template matching, the Hough transform, line adjacency graphs, character profiles, and graph grammars. Experimental results, including comparisons with commercial systems, are provided.

The traditional erosion based vectorization methods create flawed vector representations, which must then be corrected by hand. Some of these flaws are due to artifacts injected by the thinning process, while others are intrinsic to the process-perfect vectorization is not, in general, possible. The approach described here is based on looking at the complete search space, reduced by using knowledge about the image domain-that of engineering diagrams. The search space is reduced far enough that very good line and curve extractions can be performed. The results are compared against traditional vectorization in terms of the time needed by a human to repair the resulting line drawings.

While a multitude of template matching strategies have been applied to printed text recognition, the variation seen in handprinted characters generally reduces the usefulness of this technique. What is suggested in this paper is the use of scalable vector templates, which can be used to generate a template with the same scale and line width attributes as an arbitrary input character image. The best match is the template having the smallest total distance between black pixels. Multiple templates are used for each character and digits only are used as a sample data set.

Most grey-level thresholding methods produce good results in situations where the illumination gradient in the original raster image is regular and not too large. In other cases, such as a large linear change in illumination, a satisfactory bi-level image cannot be produced. If the object pixels can be identified in a variety of positions throughout the image, these can be used to construct a surface whose height is related to illumination at each pixel. This estimate can be used to produce a threshold for each pixel. The method described here uses the Shen-Castan edge detector to identify object pixels, and creates a surface using a moving least squares method that can be used to threshold the image.>

The thresholding method involves first locating objects in an image by using the intensity gradient, then noting the levels that correspond to the objects in various areas of the image, and finally using these levels as initial guesses at a threshold. This method is capable of thresholding images that have been produced in the context of variable illumination. The thresholding method, called the local intensity gradient (LIG) method, was implemented in C using a Sun4 host running UNIX. The LIG method was compared against iterative selection (IS), gray level histograms (GLHs) and two correlation based algorithms on a dozen sample images under three different illumination effects. Overall, the LIG method, while it takes significantly longer, properly thresholds a larger set of images than does any other method examined over the sample images tested.>

The usual method for specifying a fill pattern is to give a small sample of the pattern in raster form. A plotting routine then repeats this pattern over an arbitrary region. It is difficult for devices such as pen plotters to accurately reproduce pictures specified in this manner. By converting the original fill pattern to vector form, it is possible to fill arbitrary polygonal regions with a good approximation of most raster patterns.

Abstract Most programs which involve character to integer conversion fail to detect integer overflow correctly. This paper describes some of the pitfalls of not recognizing overflow, and presents a correct method for doing so in an arbitrary base for any precision.

The teaching of computer architecture at a low level is made difficult by the complexity of the real systems which are used as examples and tools. This paper describes a processor simulation system which is intended for use at the second and third year undergraduate level for teaching techniques and concepts in the implementation of instruction sets and microprogramming. The important features of this system are in the user interface, and not necessarily in the actual processor which is simulated.