Karen Brady

Former PhD Student

Contact :

Last publications in Ariana Research Group :

Models of the Unimodal and Multimodal Statistics of Adaptive Wavelet Packet Coefficients. R. Cossu and I. H. Jermyn and K. Brady and J. Zerubia. Research Report 5122, INRIA, France, February 2004. Keywords : Wavelet packet, Texture. @TECHREPORT{5122, author = {Cossu, R. and Jermyn, I. H. and Brady, K. and Zerubia, J.}, title = {Models of the Unimodal and Multimodal Statistics of Adaptive Wavelet Packet Coefficients}, year = {2004}, month = {February}, institution = {INRIA}, type = {Research Report}, number = {5122}, address = {France}, url = {http://www.inria.fr/rrrt/rr-5122.html}, pdf = {ftp://ftp.inria.fr/INRIA/publication/publi-pdf/RR/RR-5122.pdf}, ps = {ftp://ftp.inria.fr/INRIA/publication/publi-ps-gz/RR/RR-5122.ps.gz}, keyword = {Wavelet packet, Texture} } Résumé : De récents travaux ont montré que bien que les histogrammes de sous-bandes pour les coefficients d'ondelettes standards ont une forme de gaussienne généralisée, ce n'est plus vrai pour les bases de paquets d'ondelettes adaptés à une certaine texture. Trois types de statistiques sont alors observés pour les sous-bandes: gaussienne, gaussienne generalisée et dans certaines sous-bandes des histogrammes multimodaux sans mode en zéro. Dans ce rapport, nous démontrons que ces sous-bandes sont étroitement liées à la structure de la texture et sont ainsi primordiales dans les applications dans lesquelles la texture joue un rôle important. Fort de ces observations, nous étendons l'approche de modélisation de textures proposée par en incluant ces sous-bandes. Nous modifions l'hypothèse gaussienne pour inclure les gaussiennes généralisées et les mixtures de gaussiennes contraintes. Nous utilisons une méthodologie bayésienne, définissant des estimateurs MAP pour la base adaptative, pour la sélection du modèle de la sous-bande et pour les paramètres de ce modèle. Les résultats confirment l'efficacité de la méthode proposée et soulignent l'importance des sous-bandes multimodales pour la discrimination et la modélisation de textures. Abstract : In recent work, it was noted that although the subband histograms for standard wavelet coefficients take on a generalized Gaussian form, this is no longer true for wavelet packet bases adapted to a given texture. Instead, three types of subband statistics are observed: Gaussian, generalized Gaussian, and most interestingly, in some subbands, multimodal histograms with no mode at zero. As will be demonstrated in this report, these latter subbands are closely linked to the structure of the texture, and are thus likely to be important for many applications in which texture plays a role. Motivated by these observations, we extend the approach to texture modelling proposed by to include these subbands. We relax the Gaussian assumption to include generalized Gaussians and constrained Gaussian mixtures. We use a Bayesian methodology, finding MAP estimates for the adaptive basis, for subband model selection, and for subband model parameters. Results confirm the effectiveness of the proposed approach, and highlight the importance of multimodal subbands for texture discrimination and modelling. Adaptive Probabilistic Models of Wavelet Packets for the Analysis and Segmentation of Textured Remote Sensing Images. K. Brady and I. H. Jermyn and J. Zerubia. In Proc. British Machine Vision Conference (BMVC), Norwich, U. K., September 2003. Keywords : probabilistic, Adaptive, wavelet, Texture. @INPROCEEDINGS{Brady03a, author = {Brady, K. and Jermyn, I. H. and Zerubia, J.}, title = {Adaptive Probabilistic Models of Wavelet Packets for the Analysis and Segmentation of Textured Remote Sensing Images}, year = {2003}, month = {September}, booktitle = {Proc. British Machine Vision Conference (BMVC)}, address = {Norwich, U. K.}, pdf = {http://www-sop.inria.fr/members/Ian.Jermyn/publications/Brady03bmvc.pdf}, keyword = {probabilistic, Adaptive, wavelet, Texture} } Abstract : Remote sensing imagery plays an important role in many elds. It has become an invaluable tool for diverse applications ranging from cartography to ecosystem management. In many of the images processed in these types of applications, semantic entities in the scene are correlated with textures in the image. In this paper, we propose a new method of analysing such textures based on adaptive probabilistic models of wavelet packets. Our approach adapts to the principal periodicities present in the textures, and can capture long-range correlations while preserving the independence of the wavelet packet coefcients. This technique has been applied to several remote sensing images, the results of which are presented. Texture Analysis: An Adaptive Probabilistic Approach. K. Brady and I. H. Jermyn and J. Zerubia. In Proc. IEEE International Conference on Image Processing (ICIP), Barcelona, Spain, September 2003. Keywords : Adaptive, Wavelet packet, Statistics, Texture. @INPROCEEDINGS{Brady03, author = {Brady, K. and Jermyn, I. H. and Zerubia, J.}, title = {Texture Analysis: An Adaptive Probabilistic Approach}, year = {2003}, month = {September}, booktitle = {Proc. IEEE International Conference on Image Processing (ICIP)}, address = {Barcelona, Spain}, pdf = {http://www-sop.inria.fr/members/Ian.Jermyn/publications/Brady03icip.pdf}, keyword = {Adaptive, Wavelet packet, Statistics, Texture} } Abstract : Two main issues arise when working in the area of texture segmentation: the need to describe the texture accurately by capturing its underlying structure, and the need to perform analyses on the boundaries of textures. Herein, we tackle these problems within a consistent probabilistic framework. Starting from a probability distribution on the space of infinite images, we generate a distribution on arbitrary finite regions by marginalization. For a Gaussian distribution, the computational requirement of diagonalization and the modelling requirement of adaptivity together lead naturally to adaptive wavelet packet models that capture the ‘significant amplitude features’ in the Fourier domain. Undecimated versions of the wavelet packet transform are used to diagonalize the Gaussian distribution efficiently, albeit approximately. We describe the implementation and application of this approach and present results obtained on several Brodatz texture mosaics.

All publications in Ariana Research Group