Extraction de la forme et de la perspective dans des textures artificielles et des scènes naturelles par modèles corticaux
Shape and Perspective Extraction in Artificial Textures and Natural Scenes by Cortical Models
OPEN ACCESS
In this work we present a new shape from texture algorithm applied to natural scenes analysis. The originality of this approach is based on the modeling of the structure of the primary visual cortex (V1). The algorithm is able to deal with a large variety of textures presenting different types of irregularities. First to sample the amplitude spectra, we present new filters, called log-normal filters, inspired from the complex cells of V1, in replacement of the classical Gabor filters. These filters appear to be suitable for pattern analysis techniques due to their different theoretical properties, notably their radial frequency profile (adapted to the 1/f frequency profile of natural scenes) and their separability in orientation and frequency. We then use an estimation method of the local mean frequency applied to natural signals. This one does not imply the search for the adapted scale for the analysis and takes advantage of the frequencies of the used bank of filters.
Finally, from a local estimation, the orientation and shape are extracted using the geometrical properties of the perspective projection. The precision of the method is evaluated on different types of textures, both regular and irregular, and on natural scenes. The presented method allows to obtain favorably comparable results to existing best known methods with a low computational cost. Finally the model can be adapted to other applications like texture analysis, characteristic points extraction or content-based image indexation.
Résumé
Dans ce travail nous présentons un nouvel algorithme d’extraction de la forme par la texture appliqué à l’analyse des scènes naturelles. L’originalité de cette approche est basée sur la structure du cortex visuel primaire (V1) dont elle modélise les fonctions. L’algorithme est capable d’analyser une grande variété de textures présentant différents types d’irrégularités. Tout d’abord pour réaliser l’échantillonnage du spectre d’amplitude, nous proposons de nouveaux filtres, appelés filtres log-normaux, inspirés du fonctionnement des cellules complexes de l’aire V1, en remplacement des filtres de Gabor classiques. Ces filtres s’avèrent particulièrement appropriés aux techniques de reconnaissance de forme de part leurs différentes propriétés théoriques, notamment leur profil en fréquence radiale (adapté à la décroissance en 1/f des scènes naturelles) et leur séparabilité en orientation et en fréquence. Nous utilisons ensuite une méthode d’estimation de la fréquence moyenne locale appliquées sur des signaux naturels. Celle-ci ne nécessite pas la recherche d’une échelle adaptée à l’analyse et tire avantage de l’ensemble des fréquences du banc de filtres utilisé.
Finalement, à partir de l’estimation locale, l’orientation et la forme sont extraits en utilisant les propriétés géométriques de la projection perspective. La précision de la méthode est évaluée sur différents types de textures, à la fois régulières et irrégulières, et sur des scènes naturelles. La méthode présentée permet d’obtenir des résultats se comparant favorablement aux meilleures techniques existantes tout en conservant un faible coût de calcul. Enfin le modèle peut être adapté à d’autres applications telles que l’analyse de textures, l’extraction de points caractéristiques ou l’indexation d’images par le contenu.
3D perception, texture, natural scenes, log-normal filters
Mots clés
Perception 3D, texture, scènes naturelles, filtres log-normaux
[1] A. TORRALBA and A. OLIVA, Depth estimation from image structure. IEEE Pattern Analysis and Machine Intelligence, 24, No.9, 2002.
[2] B.J. SUPER and A.C. BOVIK, Planar surface orientation from texture spatial frequencies. Pattern Recognition, 28, No.5, 1995.
[3] M. CLERC and S. MALLAT, The texture gradient equation for recovering shape from texture. IEEE Trans. PAMI, 24, No.4, 2002.
[4] J. MALIK and R. ROSENHOLTZ, Computing local surface orientation and shape from texture for curved surfaces. International Journal of Computer Vision, 24, No.2, 1997.
[5] E. RIBEIRO and E.R. HANCOCK, Shape from periodic texture using the eigen vectors of local affine distortion. IEEE Trans. PAMI, 23, No.12, 2001.
[6] K. SAKAI and H. FINKEL, Spatial-frequency analysis in the perception of perspective depth. Network: Computation in Neural Systems, 8(3):335-352, 1997.
[7] A. GUERIN-DUGUE and M. ELGHADI, Shape from texture by local frequencies estimation. SCIA, Kangerlussuaq (Greenland), pages 533-544, 1999.
[8] J. GARDING and T. LINDEBERG, Direct computation of shape cues using scale-adapted spatial derivative operators. International Journal of Computer Vision, 17(2), pp. 163-191, 1996.
[9] W.S. HWANG, C.S. LU, and P.C. CHUNG, Shape from texture estimation of planar surface orientation throught the ridge surfaces of continuous wavelets transform. Trans. in Image Processing, 7, No.5, 1998.
[10] S. LELANDAIS, L. BOUTTÉ, and J. PLANTIER, Shape from texture: Local scales and vanishing line computation to improve results for macrotextures. Int. J. Image Graphics, 5(2):329-350, 2005.
[11] L. SPILLMANN and J.S. WERNER, Visual Perception : The Neurophysiological Foundations. Academic Press, Inc., 1990.
[12] W.H. BEAUDOT, The neural information in the vertebra retina: a melting pot of ideas for artificial vision. PhD thesis, tirf laboratory, Grenoble, France, 1994.
[13] D.J. HEEGER, Modeling simple-cell direction selectivity with normalized, half-squared, linear operators. Journal of Neurophysiology, 70:1885-1898, 1993.
[14] A. LI and Q. ZAIDI, Three-dimensional shape from non-homogeneous textures: carved and stretched surfaces. Journal of Vision, 4, No.10(3):860-878, 2004.
[15] J. ATICK and A. REDLICK, What does the retina know about natural scenes? Neural Computation, 4, 1992.
[16] J. HÉRAULT, A model of colour processing in the retina of vertebrates : from photoreceptors to colour opposition and colour constancy. Neurocomputing, 12:113-129, 1996.
[17] A. TORRALBA and J. HÉRAULT, An efficient neuromorphic analog network form motion estimation. IEEE Trans. on Circuits and Systems-I: Special Issue on Bio-Inspired Processors and CNNs for Vision, 46, No.2, 1999.
[18] H. KNUTSSON, C.F. WESTIN, and G. GRANLUND, Local multiscale frequency and bandwidth estimation. IEEE International Conference on Image Processing (ICIP'94), Austin, Texas, 1994.
[19] G.H. GRANLUND and H. KNUTSSON, Signal processing for computer vision. Kluwer Academic Publishers, ISBN 0-7923-9530-1., 1995.
[20] R. ROSENHOLTZ and J. MALIK, Surface orientation from texture: Isotropy or homogeneity (or both)? Vision Research, 37(16):2283-2293, 1997.
[21] D.C. KNILL, Discriminating surface slant from texture: Comparing human and ideal observers. Vision Research, 38, 1998.
[22] J. SHI and J. MALIK, Normalized cuts and image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(8):888-905, 2000.