The canonical face depth map (CFDM) is a standardized representation for storing and manipulating 3D data from human faces. Our algorithm automates the process of transforming a 3D face scan into its canonical representation, eliminating the need for hand-labeled anchor points. The presented algorithm is designed to be a robust, fully automatic preprocessor for any 3D face recognition algorithm. The experimental results presented here demonstrate that our CFDM is robust to noise and occlusion, and we show that using such a canonical representation can improve the efficiency efface recognition algorithms and reduce memory requirements. Producing the CFDM takes, on average, 0.85 seconds for 320 times 240 pixel scans, and 3.8 seconds for 640 times 480 pixel scans (using a dual AMD Opteron 275, 2.2 GHz, with 2 MB Cache, and 1 GIG RAM). The CFDM enables both 2D and 3D image processing methods - such as convolution and PCA - to be readily used for feature localization and face recognition.

The 3DID system has been developed to verify the identity of a person by matching a sensed 3D surface of the face to a face model stored during an enrollment process. Anchor point detection is performed based on shape index; then, a rigid alignment is determined between the observed and model face anchor points. The alignment is refined using a modified ICP algorithm allowing for trimming of 10% noise. Trimmed RMS error for the same person is almost always smaller than 1.3mm; whereas for different persons, it is almost always larger. Performance analysis shows that the 3DID system is fast enough (< 2 sec on a 3.2 MHz P4), reliable enough (1% EER with 1.5% reject rate), and user friendly enough (handles 30 degrees of yaw and 10 degrees of roll and pitch) to be practical in several applications. The current system provides several displays of value to human agents either in online or delayed analysis mode. An inexpensive scanner is needed for widespread use

During face-to-face collaboration people frequently monitor the other's facial expressions to determine their current state of attention, mood, and comprehension. Capturing a frontal view of the face of mobile users in multi-user collaborative environments has been a challenge for several years. A mobile social presence system has been proposed that captures two side views of the face simultaneously and generates a frontal view in real-time. The face is modeled using an active appearance model (AAM) and a mapping of the side model to the frontal model is constructed from training. Frontal views are then generated by applying this mapping to the fitted side model during collaboration. Only a few model coefficients are transmitted for the synthesized facial frames, providing a highly compressed stream. In this paper we present a performance analysis by evaluating the proposed system under limited resolution run-time conditions.

The radial mass transform (RMT) is defined to produce a rotation and translation invariant vector representation of the neighborhood structure of points. The RMT transforms 3D (or 2D) data sets into a 1D signal where m(p, r) gives the total mass (or intensity) of sensed data at distance r from the point p. A support vector machine can be trained on example signals to detect salient points in an entire image or volume. Results show the method to be effective in multiple applications. The method is computation intensive but highly parallelizable and feasible for high value data sets.

Retrieving images from a large image dataset using image content as a key is an important issue. In this paper, we present a new content-based image retrieval approach using a Wavelet transform and subband image segmentation. For the image retrieval, we first decompose the image using a Wavelet transform and adopt vector a quantization(VQ) algorithm to perform automatic segmentation based on image features such as color and texture. The wavelet transform decomposes the image into 4 subbands(LL,LH,HL,HH). Only the LL component is further decomposed until the desired depth is reached. The image segmentation is performed using the HIS color and texture features of the low pass sub-band component image. The VQ provides a transformation from the raw pixel data to a small group of homogeneous classes which are coherent in color and texture space. For managing a large image dataset, image compression is usually considered. In that sense, the segmentation of a compressed image or sub-band image is more efficient compared with using an uncompressed image since the compressed image preserves the information needed for the image segmentation task. An important aspect of the system is that using a sub-band image of the Wavelet transform can reduce the size and noise of the image. Thus, we can subsequently reduce the computational burden for the image segmentation. The experimental results of the proposed image retrieval system confirm the feasibility of our approach in retrieving accuracy and in lowering computational cost compared to using the original image.

A non-intrusive real-time program detects the face, then eyes and nose of a moving workstation user at a rate of between 10 and 30 Hertz. Based on the head pose, it determines where on the display the subject is looking. The long term goal is to provide a system for controlling a computer using head movements and gaze direction. The current program creates a base facility for other capabilities such as detecting facial gestures, creating face models, and normalizing for face recognition. A skin color model is used along with geometric knowledge about the face and weak assumptions about the lighting. Good results are reported with various subjects and conditions, including facial hair; 3D motion, and use of eyeglasses.

We present a shape-based method of indexing to model aspects from a single intensity image. Objects are assumed to be rigid. A model aspect is represented by a 2 1/2 D edgemap and the parts of the object silhouette. Part decomposition is derived from a codon representation of the object silhouette. Invariant features extracted from each part are then used to index into a hash table to generate model-aspect hypotheses. Knowledge about parts is incorporated in voting schemes to order hypotheses for efficient verification of candidate models. Verification of model-aspect hypotheses is carried out by an alignment algorithm that is robust to partial occlusion. Results of tests using 658 model aspects from 100 objects demonstrate that accurate recognition can be achieved with very few verification attempts.

A method is presented for computing the pose of rigid 3D objects with arbitrary curved surfaces. Given an input image and a candidate object model and aspect, the method will verify whether or not the object is present and if so, report pose parameters. The curvature method of Bash and Ullman is used to model points on the object rim, while stereo matching is used for internal edge points. The model allows an object edge-map to be predicted from pose parameters. Pose is computed via an iterative search for the best pose parameters. Heuristics are used so that matching can succeed in the presence of occlusion and artifact and without resetting to use of corresponding salient feature points. Bench tests and simulations show that the method almost always converges to ground truth pose parameters for a variety of objects and for a broad set of starting parameters in the same aspect.

Three-dimensional (3D) modeling based on an ensemble of multilayer self-organizing (SO) neural networks is described. Our objective for 3D modeling is to construct a representation of a 3D object shape from sensed surface points acquired from the object. Current modeling techniques can be classified into two categories: the static and the dynamic approaches, where the former grounded in computational geometry, and the latter rooted in the mechanics of elastic materials. In this paper, a neural-based dynamic modeling approach is presented. The method used is proved to converge and experimental results are shown which support its applicability to real problems.

Motivation is given for undergraduate courses involving computing with images. Such courses should overlap with, but be significantly different from, existing courses in image processing or computer vision. Images have recently become much more important as a fundamental data type, primarily because multimedia computer power has become pervasive. Future computer scientists should know how to model and manipulate images. Image computation introduces students to highly motivating applications. Moreover, the intuitive human capability for image understanding is useful for reinforcing important mathematical concepts and application projects provide significant opportunities for capstone work which integrates wide areas of undergraduate computer science.

Control of a computer workstation via face position and facial gesturing would be an important advance for people with hand or body disabilities as well as for all users. Steps toward realization of such a system are reported here. A computer system has been developed to track the eyes and the nose of a subject and to compute the direction of the face. Face direction and movement is then used to control the cursor. Test results show that the resulting system is usable, although several improvements are needed.>

This paper discusses some representation issues and challenges involved in object recognition. It is intended as a step toward assessing current object representation schemes and proposing design and evaluation criteria for future ones.>

A model-based approach is used for recognizing arterial blood vessels from MRA volumetric data. The modeling includes (1) a generalized stochastic tube model characterizing the structural properties of the vessels, and (2) a bivariate Gaussian function, modeling the expected cross sectional blood flow. This integrated model renders the recognition problem as a parameter estimation problem which is subsequently solved in a hierarchical fashion. Due to the descriptive representation for objects, a new visualization scheme for blood vessels is proposed that allows the observation of the blood flow of each cross section along a recognized vessel. Some experimental results from both synthetic data and real MRA volumes are given. The visualization of interior blood flow patterns for some vessels are also shown.>

This paper addresses the issue of tracking tubular objects, particularly blood vessels from MR images. A model-based approach is adopted. The generalized stochastic tube (GST) model is developed which is an extension of our previously proposed (1993) generalized tube (GT) model. Transitions among adjacent tubes are explicitly parameterized. Integrated with a bivariate Gaussian density function adopted to model the blood flow within cross sections, the GST model is applied to tracking blood vessels in MRA volumetric data. Experimental results on both synthetic data with different degrees of Gaussian noise and real MRA data demonstrated that simultaneously utilizing both models yields robust performance under noisy conditions.

A method of locating and tracking rigid moving objects with arbitrary curved surfaces is presented. Motion of the moving objects in a sequence of images is used to perform image segmentation and boundary extraction. The silhouette of the object model is derived by the curvature method of Basri and Ullman. The derived silhouette is then fitted to the observed silhouette to determine the object pose. Correspondence is guided by template matching, where the similarity measure is based on the minimization of the overall Euclidean distance between the derived silhouette and the observed silhouette. Bench tests and simulations confirm the viability of the approach, even when the observed silhouette is imperfect due to partial occlusion of the object or imperfect boundary extraction.>

The issue of recognizing 3D elongated objects from 2D intensity images is addressed. A tube model, locally similar to generalized cones, is developed for the class of elongated objects. A recognition strategy that combines 2D contour properties and surface shading information is used to exploit the power of the 3D model. Reliable contours provide constraints for localizing the objects of interest. The theory of optimal filters is adopted in verifying the shading of hypothesized objects. Object recognition is achieved through optimizing the signal-to-noise response with respect to model parameters. A sweeping operation is proposed as a further stage of identifying objects so that the overall performance of the system does not heavily rely on the quality of local feature detection.>

Computer operation via face orientation provides a means of communication for people with severe physical disability. The authors present a near-real-time computer system to track the eyes and the nose of a subject and compute the direction of the face. As presented by K. Ohmura et al. (1988), the direction can be computed from three natural feature points extracted from the face. The computational approach taken in this system is motivated by the practical requirements of near-real-time performance and accuracy. The approach is based on the reflection characteristic of the cornea. Any bright light source forms a twinkle in the eye that is fixed as long as the light source is fixed. Face direction detection is treated as a 2D tracking problem. The eye twinkles are extracted from the image and used to locate the nose. A final experiment shows that the face direction obtained from the three feature points is precise enough to control the mouse cursor to make selection from a menu with large light buttons.>

The minirhizotron technique has provided agricultural scientists the opportunity of observing rhizosphere activities without destroying root structures. Nonetheless, the laborious analysis of the data still prohibits its wide applications. Advanced image understanding techniques are needed to derive satisfactory descriptions of plant root networks in an efficient and robust way. The paper presents a plant root image analysis system designed as a blackboard architecture with a hierarchy of data abstractions. Important properties of plant roots are used throughout the processing and multiple sources of information are combined to resolve uncertainties in image interpretation. Experimental results from some stages of the research are given which support the overall processing scheme.>

Object wings are 2 1/2-D primitives encoding fragments of object boundary and the adjacent surfaces. The adequacy of wing representation for polygonal scenes is demonstrated in the sense that if a wing representation has been derived from the image of polygons, the spatial structure of the view can be uniquely recovered. In other words, it is possible to reconstruct all visible parts of the scene based on the given wing representation-all visible vertices the equations of all visible faces, and the completely labeled line drawing of the scene. It is shown that not only can most origami and polyhedral objects be handled but also that many views considered 'accidental' by other researchers can be handled as well.>

A computational framework is presented for constructing wing representations of general 3D rigid objects based on wing representation theory and the multiple-view paradigm. Preliminary results indicate that wing representation of objects rates reasonably well with respect to properties of locality, stability, and uniqueness; has a problem with verbosity; and ideally should possess computability. It is shown how the verbosity problem can be alleviated by using maps with lower resolutions, viewspheres with fewer viewpoints, and the hierarchic properties of both viewsphere and wing sets.>

An algorithm that acts as an oracle by estimating the number of sequential orders possible for a set of tasks and precedence constraints is presented. The basic oracle algorithm always completes in low order polynomial time even when there are an exponential number of orders, and it produces an exact answer for graphs with simple folds. The algorithm is applicable to many practical problems in CAPP and should be of value in searching for efficient sequences of operations.>

A set of 2 1/2 D primitives, called object wings, are introduced for representation and recognition of arbitrarily shaped 3D objects. There are 34 simple wings. A simple wing is defined as a triple including a pair of surface patches separated by a contour segment. Simple wings can be grouped into composite wings through nonaccidental relationships such as cotermination, symmetry, parallelism, connectivity, colinearity, and curvilinearity. Both simple wings and composite wings, together with their spatial structures, can be used to construct internal models of real-world objects. The configurations of wings possess enough information to be of use for both indexing into models and determining instance pose. Wing definitions and techniques for their detection are presented. Representation in terms of the proposed wing primitives and recognition based on the wing representations are addressed. Experimental results are given to indicate that wing primitives can be stably extracted over a broad change of scale, and that reasonable representations can be constructed.>

An intermediate representation based on combined edge and surface data is proposed to support the recognition of objects in cluttered scenes. The representation is based on the premise that local structure of surface and edge (wings) can be reliably sensed without higher-level models, while global edge, surface, and part structures cannot be. A sensing system for indoor scenes is discussed which allows fusion of an intensity image and an image where depth is coded using structured light. Surface patches are detected and coarsely typed according to only the 2D structure in the stripped image together with global knowledge of the stripe projection, while edges are detected from intensity gradients or boundaries of stripe patches. A set of rules is given for describing the 3D structure corresponding to each wing constructed in the intermediate scene representation. An extended set of rules is proposed for partitioning the set of sensed wings into subsets corresponding to solid-object parts. Reasonable results are reported when these rules are applied to several complex scenes.>

A method for 3-D surface measurement using a projected grid of light is presented. The grid line identification problem is solved using very general constraints. At the same time, 3-D surface patch solutions are developed. From the general constraints of set of geometric and topological rules are derived that are effectively applied in the computation of grid labels and hence 3-D surface solutions. A set of five algorithms has been applied on five real scenes consisting of multiple objects of arbitrary shapes. The results show that globally consistent surface solutions can be obtained rapidly with good accuracy using a single image. A small degree of ambiguity remains, but can be further reduced or removed using increased knowledge.>

Two simple methods are given for obtaining the surface shape using a projected grid. After the camera is calibrated to the 3-D workspace, the only input date needed for the computation of surface normals are grid intersect points in a single 2-D image. The first method performs nonlinear computations based on the distortion of the lengths of the grid edges and does not require a full calibration matrix. The second method requires that a full parallel projection model of the imaging is available, which enables it to compute 3-D normals using simple linear computations. The linear method performed better overall in the experiments, but both methods produced normals within 4-8 degrees of known 3-D directions. These methods appear to be superior to methods based on shape-from-shading because the results are comparable, yet the equipment setup is simpler and the processing is not very sensitive to object reflectance.>

Algorithms for matching two sets of points in a plane are given. These algorithms search in the parameter space and find the transformation parameters that can match the most points in two sets. Since an exhaustive search for the best parameters is not affordable as the number of points in the sets become large, a subset selection method is given in order to reduce the search domain. Subsets are chosen as points on the boundary of the convex hulls of the sets. The algorithms are tested on generated and real data, and their performance is compared.

This paper shows how the nondirectional structural analysis of pattern data can be performed by matching a problem reduction representation (PRR) of pattern structure with sample data, using a best-first state space search algorithm called SSS*. The end result of the matching algorithm is a tree whose nodes represent recognized structures in the data. Tip nodes of the tree structure correspond to primitives which are recognized in the raw data by curve fitting routines. The operators of the algorithm allow the tree to be constructed with a combination of top-down or bottom-up steps. The matching of the structure tree to waveform segments need not be done in a left-right sequence. Moreover ambiguous matches are pursued in a best first order by using state space search with partial parse trees as states. A software system called WAPSYS (for waveform parsing system) is described, which implements this structural analysis paradigm. Experience using WAPSYS to analyze carotid pulse waves is also discussed.

A new technique is presented for matching image features to maps or models. The technique forms all possible pairs of image features and model features which match on the basis of local evidence alone. For each possible pair of matching features the parameters of an RST (rotation, scaling, and translation) transformation are derived. Clustering in the space of all possible RST parameter sets reveals a good global transformation which matches many image features to many model features. Results with a variety of data sets are presented which demonstrate that the technique does not require sophisticated feature detection and is robust with respect to changes of image orientation and content. Examples in both cartography and object detection are given.