Spatiotemporal Gesture Segmentation

Spotting patterns of interest in an input signal is a very useful task in many different fields including medicine, bioinformatics, economics, speech recognition and computer vision. Example instances of this problem include spotting an object of interest in an image (e.g., a tumor), a pattern of interest in a time-varying signal (e.g., audio analysis), or an object of interest moving in a specific way (e.g., a human's body gesture). Traditional spotting methods, which are based on Dynamic Time Warping or hidden Markov models, use some variant of dynamic programming to register the pattern and the input while accounting for temporal variation between them. At the same time, those methods often suffer from several shortcomings: they may give meaningless solutions when input observations are unreliable or ambiguous, they require a high complexity search across the whole input signal, and they may give incorrect solutions if some patterns appear as smaller parts within other patterns. In this thesis, we develop a framework that addresses these three problems, and evaluate the framework's performance in spotting and recognizing hand gestures in video. The first contribution is a spatiotemporal matching algorithm that extends the dynamic programming formulation to accommodate multiple candidate hand detections in every video frame. The algorithm finds the best alignment between the gesture model and the input, and simultaneously locates the best candidate hand detection in every frame. This allows for a gesture to be recognized even when the hand location is highly ambiguous. The second contribution is a pruning method that uses model-specific classifiers to reject dynamic programming hypotheses with a poor match between the input and model. Pruning improves the efficiency of the spatiotemporal matching algorithm, and in some cases may improve the recognition accuracy. The pruning classifiers are learned from training data, and cross-validation is used to reduce the chance of overpruning. The third contribution is a subgesture reasoning process that models the fact that some gesture models can falsely match parts of other, longer gestures. By integrating subgesture reasoning the spotting algorithm can avoid the premature detection of a subgesture when the longer gesture is actually being performed. Subgesture relations between pairs of gestures are automatically learned from training data. The performance of the approach is evaluated on two challenging video datasets: hand-signed digits gestured by users wearing short sleeved shirts, in front of a cluttered background, and American Sign Language (ASL) utterances gestured by ASL native signers. The experiments demonstrate that the proposed method is more accurate and efficient than competing approaches. The proposed approach can be generally applied to alignment or search problems with multiple input observations, that use dynamic programming to find a solution.

Estimating Sensor Motion from Wide-Field Optical Flow on a Log-Dipolar Sensor

Log-polar image architectures, motivated by the structure of the human visual field, have long been investigated in computer vision for use in estimating motion parameters from an optical flow vector field. Practical problems with this approach have been: (i) dependence on assumed alignment of the visual and motion axes; (ii) sensitivity to occlusion form moving and stationary objects in the central visual field, where much of the numerical sensitivity is concentrated; and (iii) inaccuracy of the log-polar architecture (which is an approximation to the central 20°) for wide-field biological vision. In the present paper, we show that an algorithm based on generalization of the log-polar architecture; termed the log-dipolar sensor, provides a large improvement in performance relative to the usual log-polar sampling. Specifically, our algorithm: (i) is tolerant of large misalignmnet of the optical and motion axes; (ii) is insensitive to significant occlusion by objects of unknown motion; and (iii) represents a more correct analogy to the wide-field structure of human vision. Using the Helmholtz-Hodge decomposition to estimate the optical flow vector field on a log-dipolar sensor, we demonstrate these advantages, using synthetic optical flow maps as well as natural image sequences.

Intersubject Regularity in the Intrinsic Shape of Human V1

Previous studies have reported considerable intersubject variability in the three-dimensional geometry of the human primary visual cortex (V1). Here we demonstrate that much of this variability is due to extrinsic geometric features of the cortical folds, and that the intrinsic shape of V1 is similar across individuals. V1 was imaged in ten ex vivo human hemispheres using high-resolution (200 μm) structural magnetic resonance imaging at high field strength (7 T). Manual tracings of the stria of Gennari were used to construct a surface representation, which was computationally flattened into the plane with minimal metric distortion. The instrinsic shape of V1 was determined from the boundary of the planar representation of the stria. An ellipse provided a simple parametric shape model that was a good approximation to the boundary of flattened V1. The aspect ration of the best-fitting ellipse was found to be consistent across subject, with a mean of 1.85 and standard deviation of 0.12. Optimal rigid alignment of size-normalized V1 produced greater overlap than that achieved by previous studies using different registration methods. A shape analysis of published macaque data indicated that the intrinsic shape of macaque V1 is also stereotyped, and similar to the human V1 shape. Previoud measurements of the functional boundary of V1 in human and macaque are in close agreement with these results.

Exact Geosedics and Shortest Paths on Polyhedral Surface

We present two algorithms for computing distances along a non-convex polyhedral surface. The first algorithm computes exact minimal-geodesic distances and the second algorithm combines these distances to compute exact shortest-path distances along the surface. Both algorithms have been extended to compute the exact minimalgeodesic paths and shortest paths. These algorithms have been implemented and validated on surfaces for which the correct solutions are known, in order to verify the accuracy and to measure the run-time performance, which is cubic or less for each algorithm. The exact-distance computations carried out by these algorithms are feasible for large-scale surfaces containing tens of thousands of vertices, and are a necessary component of near-isometric surface flattening methods that accurately transform curved manifolds into flat representations.

Multi-Area Visuotopic Map Complexes in Macaque Striate and Extra-striate Cortex

We propose that a simple, closed-form mathematical expression--the Wedge-Dipole mapping--provides a concise approximation to the full-field, two-dimensional topographic structure of macaque V1, V2, and V3. A single map function, which we term a map complex, acts as a simultaneous descriptor of all three areas. Quantitative estimation of the Wedge-Dipole parameters is provided via 2DG data of central-field V1 topography and a publicly available data set of full-field macaque V1 and V2 topography. Good quantitative agreement is obtained between the data and the model presented here. The increasing importance of fMRI-based brain imaging motivates the development of more sophisticated two-dimensional models of cortical visuotopy, in contrast to the one-dimensional approximations that have been in common use. One reason is that topography has traditionally supplied an important aspect of "ground truth", or validation, for brain imaging, suggesting that further development of high-resolution fMRI will be facilitated by this data analysis. In addition, several important insights into the nature of cortical topography follows from this work. The presence of anisotropy in cortical magnification factor is shown to follow mathematically from the shared boundary conditions at the V1-V2 and V2-V3 borders, and therefore may not causally follow from the existence of columnar systems in these areas, as is widely assumed. An application of the Wedge-Dipole model to localizing aspects of visual processing to specific cortical areas--extending previous work in correlating V1 cortical magnification factor to retinal anatomy or visual psychophysics data--is briefly discussed.