Alexandra Psarrou

Key-frame extraction algorithm using entropy difference

The fast evolution of the digital video technology has opened new areas of research. The most important aspect will be to develop algorithms to perform video cataloguing, indexing and retrieval. The basic step is to find a way for video abstraction, as this will help us more for browsing a large set of video data with sufficient content representation. In this paper we present an overview of the current key-frame extraction algorithms. We propose the Entropy-Difference, an algorithm that performs spatial frame segmentation. We present evaluation of the algorithm on several video clips. Quantitative results show that the algorithm is successful in helping annotators automatically identify video key-frames

Tracking and Recognition of Face Sequences

In this work, we address the issue of encoding and recognition of face sequences that arise from continuous head movement. We detect and track a moving head before segmenting face images from on-line camera inputs. We measure temporal changes in the pattern vectors of eigenface projections of successive image frames of a face sequence and introduce the concept of “temporal signature” of a face class. We exploit two different supervised learning algorithms with feedforward and partially recurrent neural networks to learn possible temporal signatures. We discuss our experimental results and draw conclusions.

Mobile Augmented Reality for Cultural Heritage

This paper introduces an approach of using mobile Augmented Reality (mobile-AR) in cultural organisations, such as museums and archaeological sites, for information provision and enhancing the visiting experience. We demonstrate our approach by presenting a mobile-AR educational game for iPhones that has been developed for the archaeological site and the exhibition area at Sutton Hoo. This pilot aids visitors’ understanding of the site and its history via an engaging and playful game that connects the site with the British Museum where the objects that have been excavated from the site are exhibited. The paper discusses stakeholders’ requirements, the system architecture and concludes with lessons learned and future work.

2012 Special Issue: Autonomous Growing Neural Gas for applications with time constraint: Optimal parameter estimation

This paper aims to address the ability of self-organizing neural network models to manage real-time applications. Specifically, we introduce fAGNG (fast Autonomous Growing Neural Gas), a modified learning algorithm for the incremental model Growing Neural Gas (GNG) network. The Growing Neural Gas network with its attributes of growth, flexibility, rapid adaptation, and excellent quality of representation of the input space makes it a suitable model for real time applications. However, under time constraints GNG fails to produce the optimal topological map for any input data set. In contrast to existing algorithms, the proposed fAGNG algorithm introduces multiple neurons per iteration. The number of neurons inserted and input data generated is controlled autonomous and dynamically based on a priory or online learnt model. A detailed study of the topological preservation and quality of representation depending on the neural network parameter selection has been developed to find the best alternatives to represent different linear and non-linear input spaces under time restrictions or specific quality of representation requirements.

Learning Prior and Observation Augmented Density Models for Behaviour Recognition.

Recognition of human behaviours requires modeling the underlying spatial and temporal structures of their motion patterns. Such structures are intrinsically probabilistic and therefore should be modelled as stochastic processes. In this paper we introduce a framework to recognise behaviours based on both learning prior and continuous propagation of density models of behaviour patterns. Prior is learned from training sequences using hidden Markov models and density models are augmented by current visual observation.

Automatic landmarking of 2 d medical shapes using the growing neural gas network

MR Imaging techniques provide a non-invasive and accurate method for determining the ultra-structural features of human anatomy. In this study, we utilise a novel approach to segment out the ventricular system in a series of high resolution T1-weighted MR images. Our approach is based on an automated landmark extraction algorithm which automatically selects points along the contour of the ventricles from a series of 2D MRI brain images. Automated landmark extraction is accomplished through the use of the self-organising network the growing neural gas (GNG) which is able to topographically map the low dimension of the network to the high dimension of the manifold of the contour without requiring a priori knowledge of the structure of the input space. The GNG method is compared to other self-organising networks such as Kohonen and Neural Gas (NG) maps and an error metric is applied to quantify the performance of our algorithm compared to the other two.

On Utilising Template and Feature-Based Correspondence in Multi-view Appearance Models

In principle, the recovery and reconstruction of a 3D object from its 2D view projections require the parameterisation of its shape structure and surface reflectance properties. Explicit representation and recovery of such 3D information is notoriously difficult to achieve. Alternatively, a linear combination of 2D views can be used which requires the establishment of dense correspondence between views. This in general, is difficult to compute and necessarily expensive. In this paper we examine the use of affine and local feature-based transformations in establishing correspondences between very large pose variations. In doing so, we utilise a generic-view template, a generic 3D surface model and Kernel PCA for modelling shape and texture nonlinearities across views. The abilities of both approaches to reconstruct and recover faces from any 2D image are evaluated and compared.

Recognition of temporal structures: Learning prior and propagating observation augmented densities via hidden Markov states

An algorithm is described for modelling and recognising temporal structures of visual activities. The method is based on (1) learning prior probabilistic knowledge using hidden Markov models, (2) automatic temporal clustering of hidden Markov states based on expectation maximisation and (3) using observation augmented conditional density distributions to reduce the number of samples required for propagation and therefore improve recognition speed and robustness.

A generic face appearance model of shape and texture under very large pose variations from profile to profile views

Modelling the appearance of 3D objects under very large pose variations relies on recovering correspondence between local features and the texture variations across views. However, changes in object 3D pose introduce self-occlusions and cause nonlinear variations in both the shape and the texture of object appearance. In this paper we introduce a pose-invariant appearance model that utilises a generic-view shape template for alignment, Kernel PCA for modelling shape and texture nonlinearities across views of large pose variations, and a neural network for model fitting to new images.

Learning 2 D hand shapes using the topology preservation model GNG

Recovering the shape of a class of objects requires establishing correct correspondences between manually or automatically annotated landmark points. In this study, we utilise a novel approach to automatically recover the shape of hand outlines from a series of 2D training images. Automated landmark extraction is accomplished through the use of the self-organising model the growing neural gas (GNG) network which is able to learn and preserve the topological relations of a given set of input patterns without requiring a priori knowledge of the structure of the input space. To measure the quality of the mapping throughout the adaptation process we use the topographic product. Results are given for the training set of hand outlines.