A widely used approach for locating points on deformable objects in images is to generate feature response images for each point, and then to fit a shape model to these response images. We demonstrate that Random Forest regression-voting can be used to generate high quality response images quickly. Rather than using a generative or a discriminative model to evaluate each pixel, a regressor is used to cast votes for the optimal position of each point. We show that this leads to fast and accurate shape model matching when applied in the Constrained Local Model framework. We evaluate the technique in detail, and compare it with a range of commonly used alternatives across application areas: the annotation of the joints of the hands in radiographs and the detection of feature points in facial images. We show that our approach outperforms alternative techniques, achieving what we believe to be the most accurate results yet published for hand joint annotation and state-of-the-art performance for facial feature point detection.

In this paper, we present a tracking framework for capturing articulated human motions in real-time, without the need for attaching markers onto the subject's body. This is achieved by first obtaining a low dimensional representation of the training motion data, using a nonlinear dimensionality reduction technique called back-constrained GPLVM. A prior dynamics model is then learnt from this low dimensional representation by partitioning the motion sequences into elementary movements using an unsupervised EM clustering algorithm. The temporal dependencies between these elementary movements are efficiently captured by a Variable Length Markov Model. The learnt dynamics model is used to bias the propagation of candidate pose feature vectors in the low dimensional space. By combining this with an efficient volumetric reconstruction algorithm, our framework can quickly evaluate each candidate pose against image evidence captured from multiple views. We present results that show our system can accurately track complex structured activities such as ballet dancing in real-time.

For image analysis techniques to be of utility in medical diagnosis systems it is necessary to be able to perform quality control over the results they produce. Input data must conform to the assumptions within the algorithm if useful results are to be achieved. Automation of this process is essential if vision algorithms are to form components in analysis systems. In this paper we present a technique to validate the correction of field inhomogeneity in MR images. The initial intention was to use information measures to check the improvement due to correction. However, it will be shown that the standard log entropy calculation for information measurement does not have the required properties, specifically grey-level scale invariance. We present an alternative, scale-invariant information measure derived using conventional likelihood approaches, that can be applied as an absolute measure of information content. We show this technique in use for the validation of our existing coil correction method.

We describe the development of a vision system to detect natural events in a low-resolution image stream. The work involves the assessment of algorithmic design decisions to maximise detection reliability. This assessment is carried out by comparing measures and estimates made by the system under development with measures obtained independently. We show that even when these independent measures are themselves noisy, their independence can serve to guide design decisions and allow performance estimates to be made. Although presented here for one particular system design, we believe that such an approach will be applicable to other situations when an image-based system is to be used in the analysis of natural scenes in situations where a precise ground truth is not available.

A novel colour image segmentation routine, based on clustering pixels in colour space using non-parametric density estimation, is described. Although the basic methodology is well known, several important improvements to the previous work in this area are introduced. The density is estimated at a series of knot points in the colour space, and clustering is performed by hill climbing on this density function. The hill climbing is constrained such that no step crosses an intermediate Voronoid cell, ensuring that all salient clusters are detected. Most importantly, the problem of scale selection has been addressed using a statistically motivated approach, by placing the knot points according to an estimate of the noise in the original images, taking full account of error propagation in the algorithm. The algorithm has been evaluated both on synthetic data and in the context of its application in a machine vision system, specifically the calibration of velocity estimates extracted from a novel infrared sensor used in a fall detector. The application of the technique to medical images and texture recognition is also discussed.

Image subtraction is used in many areas of machine vision to identify small changes between equivalent pairs of images. Often only a small subset of the differences will be of interest. In motion analysis only those differences caused by motion are important, and differences due to other sources only serve to complicate interpretation. Simple image subtraction detects all differences regardless of their source, and is therefore problematic to use. Superior techniques, analogous to standard statistical tests, can isolate localised differences due to motion from global differences due, for example, to illumination changes. Four such techniques are described. In particular, we introduce a new non-parametric statistical measure which allows a direct probabilistic interpretation of image differences. We expect this to be applicable to a wide range of image formation processes. Its application to medical images is discussed.