British Machine Vision Association and Society for Pattern Recognition


One Day BMVA Technical Meeting in association with IEE/E4 to be held on 1st November 2000 at British Institute of Radiology, 36 Portland Place, London.

Chairpersons: Maria Petrou (Surrey); Mike Chantler (Heriot-Watt)

Image texture may result from variation in surface albedo (e.g. a printed pattern on a flat surface) or variation in surface relief (e.g. a uniformly coloured but embossed wall paper). Texture classification research has focused in the past on image texture, rather than on the underlying properties of the surface or the effect that the imaging system has on the image. However, exciting new developments that use surface information are now beginning to emerge and these are being coupled with traditional rough surface theory. Such developments can radically improve both the robustness and accuracy of texture classification systems. In addition, modern data collection techniques now yield data that refer to points distributed in volume. Variation of such data constitutes volume texture. Although such texture is difficult to perceive and visualise, it offers an additional cue for the analysis of such data with applications as diverse as medicine and geosciences.

This meeting will discuss recent advances in research concerning both these types of three dimensional texture.


10:30    Registration and coffee

10:55    Introduction and welcome, Maria Petrou, Surrey University

11:00    Photometric texture analysis, Mike Chantler, Ged McGunnigle and Jerry Wu, Heriot Watt University

11:30    Innovative techniques for the isolation and analysis of concomitant two and three-dimensional texture features, Abdul Farooq, University of the West of England

12.00    3D texture analysis for function-based engineering surface Inspection, Gui Yun Tian, University of Huddersfield

12:30    Shape-from-Texture using Local Affine Distortion, Eraldo Ribeiro and Edwin R. Hancock, University of York

13:00    Lunch

14:00    Photometric invariant statistics of rough surfaces (Provisional title) 
Maria Petrou and Maria Faraklioti (Surrey Uni)

14:30   Fractal Characterization of Hypervolume Textures, Sébastien Deguy, Université d'Auvergne, France, and Albert Benassi Université Blaise Pascal, France.

15:00    Tea

15:30    Shape from texture: homogeneity revisited, Antonio Criminisi and Andrew Zisserman (University of Oxford)

16:00    Texture Analysis Based on Affine Warping, Andrew Calway (University of Bristol)

16:30   Summary and discussion

16:40   Closing remarks and finish

REGISTRATION FORM: 1st November 2000

Please return this form to Leanne Pring, Tel 0114 272 0306, Fax 0114 272 6158 or via email to The meeting is free to members of the BMVA, or IEE but a charge of £20 is payable by non-members.  When registering please enclose a cheque for the appropriate amount made payable to "The British Machine Vision Association".

NAME: ………………………………………………………………………………….

ADDRESS: ………………………………………………………………………………….


TEL: ………………………………………………………………………………….





Title: Photometric texture analysis,
Mike Chantler, Ged McGunnigle, Jerry Wu, Heriot Watt University.

Abstract to be supplied. 

Further information can be found at:

Title: Innovative techniques for the isolation and analysis of concomitant two and three-dimensional texture features,
A. R. Farooq, M. L. Smith, L. N. Smith and P. S. Midha. Faculty of Engineering, The University of the West of England, Bristol

A review of published work relating to texture inspection, analysis and classification has been undertaken. It is evident that various definitions of 'texture' are presently in widespread use. Work undertaken here at UWE, Bristol has identified the need for a formal definition and texture taxonomy. Most researches are concerned with the analysis of 2D images, as a consequence their techniques are largely viewer-centred, and require strict environmental structuring. Further complexities include surfaces, natural or synthetic, that consist of mixed chromatic and topographic features.

The presentation will provide a brief overview of the innovative work at UWE, which is concerned, with the development of techniques for the isolation and analysis of concomitant two and three-dimensional texture features. These techniques have been based on novel adaptations of the photometric stereo method and are object-centred, relatively object pose invariant, and flexible in application. There is a need to further develop the work, to improve the existing method, develop new capabilities and widen potential applications. The talk will provide an insight into the proposed content and direction of research.

Further information can be found at:

Title: 3D texture analysis for function-based engineering surface Inspection,
Dr Gui Yun Tian, School of Engineering, University of Huddersfield, Queensgate, HD1 3DH

In this age of texture analysis much attention is paid to the surface information rather than to decide how it performs when in use. In this paper, after review of surface metrology and 3D surface texture, the paper provides application-oriented 3D-texture analysis for functional surface characterisation where the surface representing features will reflect the surface usage functions. Based on the observation, the paper will discuss morphological approach and 3D-volume texture analysis for engineering surface down to micro scales.

Title: Shape-from-Texture using Local Affine Distortion,
Eraldo Ribeiro and Edwin R. Hancock
Department of Computer Science, University of York, York Y01 5DD, UK.

This talk presents a simple approach to the recovery of dense orientation estimates for curved textured surfaces. We make two contributions. Firstly, we show how pairs of spectral peaks can be used to make direct estimates of the slant and tilt angles for local tangent planes to the textured surface. We commence by computing the affine distortion matrices for pairs of corresponding spectral peaks. The key theoretical contribution is to show that the directions of the eigenvectors of the affine distortion matrices can be used to estimate local slant and tilt angles. In particular, the leading eigenvector points in the tilt direction. Although not as geometrically transparent, the direction of the second eigenvector can be used to estimate the slant direction. The main practical benefit furnished by our analysis is that it allows us to estimate the orientation angles in closed form without recourse to numerical optimisation. Based on these theoretical properties we present an algorithm for the analysis of regularly textured curved surfaces. 

The second contribution of the paper is to show how initial orientation estimates delivered by the eigen-analysis can be refined using a process of robust smoothing. We apply the method to a variety of real-world and synthetic imagery. We show that the new shape-from-texture method can reliably estimate surface topography.

Our results suggest that our new treatment of the coordinate frame ambiguity problem in bundle adjustment achieves faster and more stable convergence than existing methods.

Further information can be found at:

Title: Photometric invariant statistics of rough surfaces (Provisional title), 
Maria Petrou and Maria Faraklioti, 
Department of Electronic and Electrical Engineering, University of Surrey, Guildford, Surrey, GU2 7XH.

Abstract to be supplied.

Title: Fractal Characterization of Hypervolume Textures
Sébastien Deguy(1) and Albert Benassi(2)

(1) LLAIC1, Laboratoire de Logique, Algorithmique et Informatique de Clermont 1, Université d'Auvergne IUT Département Informatique BP 86, 63172 Aubière Cedex, France, e-mail :

(2) LaMP, Laboratoire de Météorologie Physique, Université Blaise Pascal, Pôle Physique, Complexe scientifique des Cézeaux, 63177 Aubière Cedex, France, e-mail :

We introduce an n-D extension of texture fractal analysis, based on an estimation of the Hurst parameter, linearly related to the fractal dimension. This parameter represents the power law of the fractional Brownian motion, the most simple random fractal process. The texture is then seen as a realization of this process, and is characterized through an estimation of the Hurst parameter.

We firstly present a statistical estimator of the Hurst parameter for 1-D textures. We therefore propose an extension of this estimator to the n-D case, and present a simple algorithm for "hypervolume textures" fractal analysis. This algorithm is based on a multi-scale Laplacian computing. Some possible n-D multi-scale Laplacians are then presented.

We therefore validate the approach, computing the estimation for synthesized 2- and 3-D fractional Brownian motions with varying Hurst parameter values, verifying the effectiveness of the algorithm, and discussing the respective interest of the Laplacians.

The second part of the presentation concerns preliminary works dealing with in vitro cells analysis. The cells aggregate volume texture is acquired using laser confocal microscopy. This method is non destructive, and based on virtual optical cuts in the observed object. The only fluorescence image, emitted in the plane, is recorded. This succession of planes therefore constitutes the volume data.

Through a volume texture analysis, the aim is to characterize the cells growth (known as fractal), and to segment cells in 3-D, in order to extract size and repartition informations. Preliminary results, outcoming of the proposed fractal analysis of this volume texture and of 3-D segmentation, are therefore proposed.

Texture Analysis Based on Affine Warping

Andrew Calway
University of Bristol

Local affine warping provides a powerful computational model in 2-D texture analysis. The classic structural model of texture based on the repetition of a basic structural element - the texton - can be enhanced by allowing local coordinate transformations (warp) of the texton at each location, reflecting either variation in the texture pattern or the structure of the underlying 3-D surface. Similarly, the perspective distortion of projected texture caused by a moving surface or camera can also be approximated well by a piecewise linear (affine) field, enabling the task of estimating motion or region correspondence to be simplified. The purpose of this talk is to give an overview of work based on such models carried out by the author over the past 5 years. This has centred around the development of a robust affine correspondence algorithm based on multiresolution frequency domain methods and its application to areas such as texture synthesis, shape-from-texture, 2-D motion estimation, and 3-D structure from 2-D motion. The talk will concentrate on the underlying models adopted and present examples from the various applications. Further details are available from relevant publications at: