We address the problem of anomaly detection in machine perception. The concept of domain anomaly is introduced as distinct from the conventional notion of anomaly used in the literature. We propose a unified framework for anomaly detection which exposes the multifaceted nature of anomalies and suggest effective mechanisms for identifying and distinguishing each facet as instruments for domain anomaly detection. The framework draws on the Bayesian probabilistic reasoning apparatus which clearly defines concepts such as outlier, noise, distribution drift, novelty detection (object, object primitive), rare events, and unexpected events. Based on these concepts we provide a taxonomy of domain anomaly events. One of the mechanisms helping to pinpoint the nature of anomaly is based on detecting incongruence between contextual and noncontextual sensor(y) data interpretation. The proposed methodology has wide applicability. It underpins in a unified way the anomaly detection applications found in the literature. To illustrate some of its distinguishing features, in here the domain anomaly detection methodology is applied to the problem of anomaly detection for a video annotation system.

The field of Action Recognition has seen a large increase in activity in recent years. Much of the progress has been through incorporating ideas from single-frame object recognition and adapting them for temporal-based action recognition. Inspired by the success of interest points in the 2D spatial domain, their 3D (space-time) counterparts typically form the basic components used to describe actions, and in action recognition the features used are often engineered to fire sparsely. This is to ensure that the problem is tractable; however, this can sacrifice recognition accuracy as it cannot be assumed that the optimum features in terms of class discrimination are obtained from this approach. In contrast, we propose to initially use an overcomplete set of simple 2D corners in both space and time. These are grouped spatially and temporally using a hierarchical process, with an increasing search area. At each stage of the hierarchy, the most distinctive and descriptive features are learned efficiently through data mining. This allows large amounts of data to be searched for frequently reoccurring patterns of features. At each level of the hierarchy, the mined compound features become more complex, discriminative, and sparse. This results in fast, accurate recognition with real-time performance on high-resolution video. As the compound features are constructed and selected based upon their ability to discriminate, their speed and accuracy increase at each level of the hierarchy. The approach is tested on four state-of-the-art data sets, the popular KTH data set to provide a comparison with other state-of-the-art approaches, the Multi-KTH data set to illustrate performance at simultaneous multiaction classification, despite no explicit localization information provided during training. Finally, the recent Hollywood and Hollywood2 data sets provide challenging complex actions taken from commercial movie sequences. For all four data sets, the proposed hierarchical approach outperforms all other methods reported thus far in the literature and can achieve real-time operation.

Within the field of action recognition, features and descriptors are often engineered to be sparse and invariant to transformation. While sparsity makes the problem tractable, it is not necessarily optimal in terms of class separability and classification. This paper proposes a novel approach that uses very dense corner features that are spatially and temporally grouped in a hierarchical process to produce an overcomplete compound feature set. Frequently reoccurring patterns of features are then found through data mining, designed for use with large data sets. The novel use of the hierarchical classifier allows real time operation while the approach is demonstrated to handle camera motion, scale, human appearance variations, occlusions and background clutter. The performance of classification, outperforms other state-of-the-art action recognition algorithms on the three datasets; KTH, multi-KTH, and Hollywood. Multiple action localisation is performed, though no groundtruth localisation data is required, using only weak supervision of class labels for each training sequence. The Hollywood dataset contain complex realistic actions from movies, the approach outperforms the published accuracy on this dataset and also achieves real time performance.

The paper presents a fuzzy chamfer distance and its probabilistic formulation for edge-based visual tracking. First, connections of the chamfer distance and the Hausdorff distance with fuzzy objective functions for clustering are shown using a reformulation theorem. A fuzzy chamfer distance (FCD) based on fuzzy objective functions and a probabilistic formulation of the fuzzy chamfer distance (PFCD) based on data association methods are then presented for tracking, which can all be regarded as reformulated fuzzy objective functions and minimized with iterative algorithms. Results on challenging sequences demonstrate the performance of the proposed tracking method.

The ever growing need for improved security, surveillance and identity protection, calls for the creation of evermore reliable and robust face recognition technology that is scalable and can be deployed in all kinds of environments without compromising its effectiveness. In this paper we study the impact that pose correction has on the performance of 2D face recognition. To measure the effect, we use a state of the art 2D recognition algorithm. The pose correction is performed by means of 3D morphable model. Our results on the non frontal XM2VTS database showed that pose correction can improve recognition rates up to 30%.

Deformable surface fitting methods have been widely used to establish dense correspondence across different 3D objects of the same class. Dense correspondence is a critical step in constructing morphable face models for face recognition. In this paper a mainstream method for constructing dense correspondences is evaluated on 912 3D face scans from the Face Recognition Grand Challenge FRGC V1 database. A number of modifications to the standard deformable surface approach are introduced to overcome limitations identified in the evaluation. Proposed modifications include multi-resolution fitting, adaptive correspondence search range and enforcing symmetry constraints. The modified deformable surface approach is validated on the 912 FRGC 3D face scans and is shown to overcome limitations of the standard approach which resulted in gross fitting errors. The modified approach halves the rms fitting error with 98% of points within 0.5mm of their true position compared to 67% with the standard approach.

The majority of camera calibration methods, including the Gold Standard algorithm, use point-based information and simultaneously estimate all calibration parameters. In contrast, we propose a novel calibration method that exploits line orientation information and decouples the problem into two simpler stages. We formulate the problem as minimization of the lateral displacement between single projected image lines and their vanishing points. Unlike previous vanishing point methods, parallel line pairs are not required. Additionally, the invariance properties of vanishing points mean that multiple images related by pure translation can be used to increase the calibration data set size without increasing the number of estimated parameters. We compare this method with vanishing point methods and the Gold Standard algorithm and demonstrate that it has comparable performance.

We consider the problem of decoupling translation and rotation for a collec-tion of 3D data related to 2D images by a projection. The main contribution is to show that equations describing image formation can be decoupled to form a component independent of translation. The decomposition is based on the invariance properties of the projection of points at infinity; if image formation is expressed in terms of points at infinity, general motions can be reduced to pure rotations. Contrary to other methods based on vanishing points, our approach does not require parallel directions to be present in the scene. We use the invariance property to simplify camera calibration equa-tions. We consider three cases for a known calibration object: full calibration, pose estimation and internal calibration by pure translation. Experiments on synthetic and real data show that the decomposition can obtain similar re-sults to a full parameter search. For pure translation, the decomposition can be effectively used to obtain more accurate parameters, without increasing the dimensionality of the problem. 1

This paper addresses the problem of estimating the shape of an actor in a multi-camera studio for arbitrarily positioned cameras and arbitrary human pose. We adopt a seamless articulated mesh model and introduce a novel shape matching technique to automatically transform the projected shape of the model to match multiple captured image silhouettes. Our approach treats the projected mesh as a deformable model and constrains the model to follow a smooth shape transformation to match each image silhouette. Multiple 2D transformations are integrated in 3D to update the shape of the model to match the actor. We assess the technique using virtual views generated for 3D scanned human data-sets and present preliminary results in a studio. 1

This paper presents a layered animation framework which uses displacement maps for efficient representation and animation of highly detailed surfaces. The model consists of three layers: a skeleton; low-resolution control model; and a displacement map image. The novel aspects of this approach are an automatic closed-form solution for displacement map generation and animation of the layered displacement map model. This approach provides an efficient representation of complex geometry which allows realistic deformable animation with multiple levels-of-detail.

A study of several methods for combining information from two classifiers in a system for the recognition of handwritten text is presented. The system uses two hidden Markov models (HMMs) per character to model columns and rows of pixels in the character image. We show that the best method of combining the results from the vertical and horizontal classifiers is simply to multiply the probabilities produced by the two methods. This approach outperforms more complicated classifier combination strategies such as the behaviour-knowledge space (BKS) method.

We apply an HMM-based text recognition system to the recognition of handwritten digit strings of unknown length. The algorithm is tailored to the input data by controlling the maximum number of levels searched by the level building (LB) search algorithm. We demonstrate that setting this parameter according to the pixel length of the observation sequence, rather than using a fixed value for all input data, results in a faster and more accurate system. Best results were achieved by setting the maximum number of levels to twice the estimated number of characters in the input string. We also describe experiments which show the potential for further improvement by using an adaptive termination criterion in the LB search.

We present a new method for the recognition of polyhedral objects from 2D images based on geometric hashing. Rather than the point-based approach of previous geometric hashing implementations, which tend to be rather sensitive to image noise and spurious data, our method is based on triples of connected edges. As well as improving the robustness of the system, the use of higher level feature groupings results in a very efficient specialisation of the geometric hashing paradigm. Theoretical analyses of the ForeSight method show that it is more than ten times as fast as a comparable point-based geometric hashing implementation, while using only one-quarter of the memory. These results were confirmed by practical experiments on a database of 50 real images, in which the recognition rate achieved by ForeSight approached twice that of the conventional method.

A new surface based approach to implicit surface polygonisation is introduced. This is applied to the reconstruction of 3D surface models of complex objects from multiple range images. Geometric fusion of multiple range images into an implicit surface representation was presented in previous work. This paper introduces an efficient algorithm to reconstruct a triangulated model of a manifold implicit surface, a local 3D constraint is derived which defines the Delaunay surface triangulation of a set of points on a manifold surface in 3D space. The 'marching triangles' algorithm uses the local 3D constraint to reconstruct a Delaunay triangulation of an arbitrary topology manifold surface. Computational and representational costs are both a factor of 3-5 lower than previous volumetric approaches such as marching cubes.

Rumours of the death of the problem of machine-printed text recognition have been greatly exaggerated. Reported results can be good enough to lead one to believe that this is a "solved problem". Closer analysis reveals test data that is often limited in its range of fonts and point sizes. Worse still, results are commonly quoted for noise-free images, ignoring the problems of recognising "real" documents such as faxes. Various methods have been proposed for modelling characters with Hidden Markov Models. The authors, amongst others, have suggested representing a character by analysing the pixel pattern in columns of its image, and linking sequential column patterns together with a HMM. In this paper we propose a method of quantising the patterns by means of a Shift Invariant Hamming Distance. A full experimental evaluation (45 fonts, 5 point sizes) in typical noise results in a recognition accuracy of 99% in the top-3 choices, and 94% top-choice for the best font. The method has a significant advantage in recognising noisy word images, due to classification being achieved without a prior segmentation of the word into characters.

In this paper, a novel method is proposed for the recognition of noisy printed characters. The method is based on the representation of the shape of a character by two Hidden Markov Models. Recognition is achieved by scoring these models against the test pattern and combining the results. The method has been evaluated using Baird's noise model, producing a peak performance of 99.5% on the test set in the presence of near-minimal noise. The method generalises to recognise characters with noise levels greater than those included in the training set, and an investigation of the top-k performance suggests that a much higher recognition rate could be achieved on language text using a context driven word recogniser.

Deformable surfaces have many applications in surface reconstruction, tracking and segmentation of range or volumetric data. Many existing deformable surfaces connect control points in a predefined and inflexible way. This means that the surface topology is fixed in advance, and also imposes severe limitations on how a surface can be described. For example a rectangular grid of control points cannot be evenly distributed over a sphere, and singularities occur at the poles. In this paper we introduce a new (G 1 continuous) deformable surface. In contrast to other methods this method can represent a surface of arbitrary topology, and do so in an efficient way. The method is based on a generalization of biquadratic B-splines, and has a comparable computational cost to methods based on traditional tensor product B-splines. 1

Range images may be used for a variety of applications in object recognition, inspection and reverse engineering. In many of these applications it is important to obtain good estimates of the local surface curvature. Good curvature estimates require good derivative estimates, but the estimation of derivatives from sampled data is highly susceptible to noise. In this paper, we introduce a new way of characterizing range data by a single parameter. From this characterization we show how to make an optimal choice of whatever parameters there are in a particular derivative estimation method, and obtain an estimate of the error one might expect. Finally, we show how the analysis may be applied to measuring derivatives on a cylinder.

This paper addresses the problem of calibrating a camera mounted on a robot arm. The objective is to estimate the camera's intrinsic and extrinsic parameters. These include the relative position and orientation of camera with respect to robot base as well as the relative position and orientation of the camera with respect to a pre-defined world frame. A calibration object with a known 3D shape is used together with two known movements of the robot. A method is presented to find calibration parameters within an opti-misation framework. This method differs from existing methods in that 1) it fully exploits information from different displacements of the camera to produce an optimal calibration estimate, and 2) it uses an evolutionary algo-rithm to attain the optimal solution. Experimental results on both synthetic and real data are presented. 1

Reliable curvature estimation is an important goal in image analysis to provide viewpoint independent cues for shape classification. This paper presents a model of the relationship between the variance of curvature estimates and the image noise. Agreement to within 10% is obtained for 3D range data. Previous models have only provided qualitative agreement with experimental observations. A perturbation error analysis is performed on the local least square surface fitting algorithm which is commonly used to obtain partial derivative estimates in the presence of noise.

In this paper a formal, quantitative approach to designing optimum Hough transform (HT) algorithms is proposed. This approach takes the view that a HT is a hypothesis testing method. Each sample in the HT array implements a test to determine whether a curve with the given parameters fits the edge point data. This view allows the performance of HT algorithms to be quantified. The power function, which gives the probability of rejection as a function of the underlying parametric distribution of data points, is shown to be the fundamentally important characteristic of HT behaviour. Attempting to make the power function narrow is a formal approach to optimizing HT performance. To illustrate how this framework is useful the particular problem of line detection is discussed in detail. It is shown that the hypothesis testing framework leads to a redefinition of the HT in which the values are a measure of the distribution of points around a curve rather than the number of points on a curve. This change dramatically improves the sensitivity of the method to small structures. The solution to many HT design problems can be posed within the framework, including optimal quantizations and optimum sampling of the parameter space. In this paper the authors consider the design of optimum I-D filters, which can be used to sharpen the peak structure in parameter space. Results on several real images illustrate the improvements obtained.>

In this paper 3D motion estimation using an active stereo vision paradigm is addressed. The objective is to measure camera rotations of a stereo head to estimate the 3D motion of a single moving object in a scene. The method uses stereo camera fixation of junctions. The vertex of a junction is selected as the fixation point. Vertex position and junction orientations are recovered in the fixated views and used with known camera motion to estimate 3D rigid object motion. Experiments are reported for real data and promising results are obtained. 1

We investigate the use of Robust Estimation in an application requiring the accurate location of the centres of circular objects in an image. A common approach used throughout computer vision for extracting shape information from a data set is to fit a feature model using the Least Squares method. The well known sensitivity of this method to outliers is traditionally accommodated by outlier rejection methods. These usually consist of heuristic applications of model templates or data trimming. Robust Estimation offers a theoretical framework for assessing such rejection schemes, and more importantly, provides an approach to parameter estimation in contaminated data distributions capable of greater accuracy.

A formal, quantitative approach to designing optimal Hough transform (HT) algorithms is proposed. This approach takes the view that a HT is a hypothesis testing method. This view allows the performance of HT algorithms to be quantified. The power function, which gives the probability of rejection as a function of the underlying parametric distribution of data points, is shown to be the fundamentally important characteristic of HT behavior. Attempting to make the power function narrow is a formal approach to optimizing HT performance. To illustrate how this framework is useful the particular problem of line detection is discussed. It is shown that the hypothesis testing framework leads to a redefinition of HT in which the values are a measure of the distribution of points around a curve rather than the number of points on a curve. The solution to many HT design problems can be posed within the framework, including optimal quantizations and optimum sampling of the parameter space. Consideration is given to the design of optimum 1-D filters which can be used to sharpen the peak structure in parameter space. Results on several real images illustrate the improvements obtained.>

An efficient method for finding straight lines in edge maps is described. The algorithm is based on a pyramid structure. At the bottom level of the pyramid, short line segments are detected by applying a Hough transform to small subimages. The algorithm proceeds, bottom up, from this low-level description by grouping line segments within the local neighborhoods into longer lines. Line segments which have local support propagate up the hierarchy and take part in grouping at higher levels. The length of a line determines approximately the level in the pyramid to which it propagates, and a hierarchical description of the line segments in a scene is obtained. The line-segment-grouping process is also based on a Hough transform. The method is quite efficient and has an attractive architecture which is suitable for parallel implementation.>

We introduce the Adaptive Hough Transform, AHT, as an efficient way of implementing the Hough Transform, HT, method for the detection of 2-D shapes. The AHT uses a small accumulator array and the idea of a flexible iterative "coarse to fine" accumulation and search strategy to identify significant peaks in the Hough parameter spaces. The method is substantially superior to the standard HT implementation in both storage and computational requirements. In this correspondence we illustrate the ideas of the AHT by tackling the problem of identifying linear and circular segments in images by searching for clusters of evidence in 2-D parameter spaces. We show that the method is robust to the addition of extraneous noise and can be used to analyze complex images containing more than one shape.

The threshold selection method of N. Otsu (ibid., vol.SMC-9, p.62-6, 1979) is shown to break down for a certain range of object-to-background pixel population ratios. Modifications to Otsu's method are proposed to overcome some of its limitations. The findings are also relevant to the closely related methods of T. Ridler (1978) and H.J. Trussell (1979).