FREE ELECTRONIC LIBRARY - Dissertations, online materials

Pages:   || 2 |

«Rutb Rosenboltz· XeroxPARC 3333 Coyote Hill Rd. Palo Alto, CA 94304 Abstract We suggest a working definition of texture: Texture is stuff that is ...»

-- [ Page 1 ] --

General-purpose localization of textured

• •

Image regions

Rutb Rosenboltz·


3333 Coyote Hill Rd.

Palo Alto, CA 94304


We suggest a working definition of texture: Texture is stuff that is

more compactly represented by its statistics than by specifying the

configuration of its parts. This definition suggests that to fmd

texture we look for outliers to the local statistics, and label as

texture the regions with no outliers. We present a method, based upon this idea, for labeling points in natural scenes as belonging to texture regions, while simultaneously allowing us to label low- level, bottom-up cues for visual attention. This method is based upon recent psychophysics results on processing of texture and popout.



In a number of problems in computer VlSlon and image processing, one must distinguish between image regions that correspond to objects and those which correspond to texture, and perform different processing depending upon the type of region. Current computer vision algorithms assume one magically knows this region labeling. But what is texture? We have the notion that texture involves a pattern that is somehow homogeneous, or in which signal changes are "too complex" to describe, so that aggregate properties must be used instead (Saund, 1998). There is by no means a firm division between texture and objects; rather, the characterization often depends upon the scale of interest (Saund, 1998).

• Email: rruth@parc.xerox.com 818 R. Rosenholtz Ideally the defmition of texture should probably depend upon the application. We investigate a definition that we believe will be of fairly general utility: Texture is stuff that seems to belong to the local statistics. We propose extracting several texture features, at several different scales, and labeling as texture those regions whose feature values are likely to have come from the local distribution.

Outliers to the local statistics tend to draw our attention (Rosenholtz, 1997, 1998).

The phenomenon is often referred to as "popout." Thus while labeling (locally) statistically homogeneous regions as texture, we can simultaneously highlight salient outliers to the local statistics. Our revised defmition is that texture is the absence of popout.

In Section 2, we discuss previous work in both human perception and in fmding texture and regions of interest in an image. In Section 3, we describe our method.

We present and discuss results on a number of real images in Section 4.


See (Wolfe, 1998) for a review of the visual search literature. Popout is typically studied using simple displays, in which an experimental subject searches for the unusual, target item, among the other, distractor items. One typically attempts to judge the "saliency," or degree to which the target pops out, by studying the efficiency of search for that item. Typically popout is modeled by a relatively lowlevel operator, which operates independently on a number of basic features of the image, including orientation, contrast/color, depth, and motion. In this paper, we look only at the features of contrast and orientation.

Within the image-processing field, much of the work in fmding texture has defmed as texture any region with a high luminance variance, e.g. Vaisey & Gersho (1992).

Unfortunately, the luminance variance in a region containing an edge can be as high as that in a textured region. Won & Park (1997) use model fitting to detect image blocks containing an edge, and then label blocks with high variance as containing texture.

Recently, several computer vision researchers have also tackled this problem.

Leung & Malik (1996) found regions of completely deterministic texture. Other researchers have used the defmition that if the luminance goes up and then down again (or vice versa) it's texture (Forsyth et aI, 1996). However, this method will treat lines as if they were texture. Also, with no notion of similarity within a texture (also lacking in the image-processing work), one would mark a "fault" in a texture as belonging to that texture. This would be unacceptable for a texture synthesis application, in which a routine that tried to synthesize such a texture would most likely fail to reproduce the (highly visible) fault. More recently, Shi and Malik (1998) presented a method for segmenting images based upon texture features. Their method performs extremely well at the segmentation task, dividing an image into regions with internal similarity that is high compared to the similarity across regions. However, it is difficult to compare with their results, since they do not explicitly label a subset of the resulting regions as texture. Furthermore, this method may also tend to mark a "fault" in a texture as belonging to that texture.

This is both because the method is biased against separating out small regions, and because the grouping of a patch with one region depends as much upon the difference between that patch and other regions as it does upon the similarity between the patch and the given region.

Very little computer vision work has been done on attentional cues. Milanese et al (1993) found salient image regions using both top-down information and a bottomup "conspicuity" operator, which marks a local region as more salient the greater the General-Purpose Localization o/Textured Image Regions 819 difference between a local feature value and the mean feature value in the surrounding region. However. for the same difference in means. a local region is less salient when there is a greater variance in the feature values in the surrounding region (Duncan & Humphreys. 1989; Rosenholtz. 1997). We use as our saliency measure a test for outliers to the local distribution. This captures. in many cases. the dependence of saliency on difference between a given feature value and the local mean. relative to the local standard deviation. We will discuss our saliency measure in greater detail in the following section.


We compute multiresolution feature maps for orientation and contrast. and then look for outliers in the local orientation and contrast statistics. We do this by fast creating a 3-level Gaussian pyramid representation of the image. To extract contrast. we filter the pyramid with a difference of circularly symmetric Gaussians.

The response of these filters will oscillate. even in a region with constant-contrast texture (e.g. a sinewave pattern). We approximate a computation of the maximum response of these filters over a small region by fast squaring the filter responses.

and then filtering the contrast energy with an appropriate Gaussian. Finally. we threshold the contrast to eliminate low-contrast regions ("flat" texture). These thresholds (one for each scale) were set by examining the visibility of sinewave patterns of various spatial frequencies.

We compute orientation in a simple and biologically plausible way. using Bergen &

Landy's (1991) "back pocket model" for low-level computations:

1. Filter the pyramid with horizontal. vertical. and ±45° oriented Gaussian second derivatives.

2. Compute opponent energy by squaring the filter outputs. pooling them over a region 4 times the scale of the second derivative filters. and subtracting the vertical from the horizontal response and the +45 0 from the _45 0 response.

3. Normalize the opponent energy at each scale by dividing by the total energy in the 4 orientation energy bands at that scale.

The result is two images at each scale of the pyramid. To a good approximation. in regions which are strongly oriented. these images represent kcos(26) and ksin(26).

where 6 is the local orientation at that scale. and k is a value between 0 and 1 which is related to the local orientation specificity. Orientation estimates from points with low specificity tend to be very noisy. In images of white noise. 80% of the estimates of k fall below 0.5. therefore with 80% confidence. an orientation specificity of k0.5 did not occur due to chance. We use this value to threshold out orientation estimates with low "orientedness.·· We then estimate D, the local feature distribution, for each feature and scale, using the method of Parzen windows. The blurring of the distribution estimate by the Parzen window mimics uncertainty in estimates of feature values by the visual system. We collect statistics over a local integration region. For texture processing.

the size of this region is ind.ependent of viewing distance, and is roughly lOS in diameter, where S is the support of the Gaussian 2nd derivative filters used to extract the texture features (Kingdom & Keeble, 1997; Kingdom et ai, 1995).

We next compute a non-parametric measure of saliency:

–  –  –

Note that if D were Gaussian N(~,a2), this simplifies to (X_tt)2 (2) '1a 2 which should be compared to the standard parametric test for outliers, which uses the measure (x - tt)/a. Our saliency measure is essentially a more general, nonparametric form of this measure (i.e. it does not assume a Gaussian distribution).

Points with saliency less than 0.5 are labeled as candidate texture points. If D were Gaussian, this would correspond to feature estimates within one standard deviation of the mean. Points with saliency greater than 3.1 are labeled as candidates for bottom-up attentional cues. If D were Gaussian, this would correspond to feature estimates more than 2.50 from the mean, a standard parametric test for outliers.

One could, of course, keep the raw saliency values, as a measure of the likelihood that a region contained texture, rather than setting a hard threshold. We use a hard threshold in our examples to better display the results. Both the texture images and the region of interest images are median-filtered to remove extraneous points.


Figure 3 shows several example images. Figures 2, 3, and 4 show texture found at each scale of processing. The striped and checkered patterns represent oriented and homogeneous contrast texture, respectively. The absence of an image in any of these figures means that no texture of the given type was found in that image at the given scale. Note that we perform no segmentation of one texture from another.

For the building image, the algorithm labeled bricks and window panes as fme-scale texture, and windows and shutters as coarser-scale texture. The leopard skin and low-frequency stripes in the lower right comer of the leopard image were correctly labeled as texture. In the desk image, the "wood" texture was correctly identified.

The regular pattern of windows were marked as texture in the hotel image. In the house image, the wood siding, trees, and part of the grass were labeled as texture (much of the grass was low contrast and labeled as "flat" texture). One of the bushes is correctly identified as having coarser texture than the other has. In the lighthouse image, the house sans window, fence, and tower were marked, as well as a low-frequency oriented pattern in the clouds.

Figure 5 shows the regions of interest that were found (the striped and plaid patterns here have no meaning but were chosen for maximum visibility). Most complex natural scenes had few interesting low-level attentional areas. In the lighthouse image, the life preserver is marked. In the hotel, curved or unusual angular windows are identified as attentional cues, as well as the top of the building. Both of these results are in agreement with psychophysical results showing that observers quickly identify curved or bent lines among straight lines (reviewed in Wolfe, 1998). The simpler desk scene yields more intuitive results, with each of the 3 objects labeled, as well as the phone cord.

Bottom-up attentional cues are outliers to the local distribution of features, and we have suggested that texture is the absence of such outliers. This definition captures some of the intuition that texture is homogeneous and statistical in nature. We presented a method for fmding contrast and orientation outliers, and results both on localizing texture and on finding popout in natural images. For the simple desk image, the algorithm highlights salient regions that correspond to our notions of the important objects in the scene. On complicated natural scenes, its results are less intuitive; suggesting that search in natural scenes makes use of higher-level 821 General-Purpose Localization o/Textured Image Regions processing such as grouping into objects. This result should not be terribly surprising, but serves as a useful check on simple low-level models of visual attention. The algorithm does a good job of identifying textured regions at a number of different scales, with the results perhaps more intuitive at finer scales.

Acknowledgments This work was partially supported by an NRC postdoctoral award at NASA Ames.

Many thanks to David Marimont and Eric Saund for useful discussions.

References J. R. Bergen and M. S. Landy (1991), "Computational modeling of visual texture segmentation," Computational Models of Visual Processing, Landy and Movshon (eds.), pp. 252-271, MIT Press, Cambridge, MA.

J. Duncan and G. Humphreys (1989), "Visual search and stimulus similarity," Psych. Review 96, pp. 433-458.

D. Forsyth, J. Malik, M. Fleck, H. Greenspan, T. Leung, S. Belongie, C. Carson, and C. Bregler (1996), "Finding pictures of objects in collections of images," ECCV Workshop on Object Representation, Cambridge.

F. A. A. Kingdom, D. Keeble, D., and B. Moulden (1995), "Sensitivity to orientation modulation in micropattern-based textures," Vis. Res. 35, 1, pp. 79-91.

F. A. A. Kingdom and D. Keeble (1997), "The mechanism for scale invariance in orientation-defined textures." Invest. Ophthal. and Vis. Sci. (Suppl.) 38, 4, p. 636.

T. K. Leung and J. Malik (1996), "Detecting, localizing, and grouping repeated scene elements from an image," Proc. 4th European Con! On Computer Vision, 1064, 1, pp. 546-555, Springer-Verlag, Cambridge.

R. Milanese, H. Wechsler, S. Gil, J. -M. Bost, and T. Pun (1993), "Integration of bottom-up and top-down cues for visual attention using non-linear relaxation," Proc. IEEE CVPR, pp. 781-785, IEEE Computer Society Press, Seattle.

Pages:   || 2 |

Similar works:

«1 Pierrette Meisch, criminologue, Baptiste Antoine, licencié en licenciée en santé publique, formée sciences psychologiques, formé en en approche familiale systémique et médiation, Centre de Consultation et en médiation, directrice à la de Médiation Familiale de la Fondation Pro Familia Fondation Pro Familia Fondation Pro Familia A la recherche des significations des violences en milieu scolaire touchant les enfants « souffre-douleur » Abstract. L’article propose une réflexion...»

«What Indoor Cats Need To enrich the lives of indoor cats, we have developed this resource checklist; and some suggestions for making changes. Informed Owners As an owner, one of the most important things you can do for you cat is to educate yourself about feline idiosyncrasies. These resources will help you do just that. Books From the Cat's Point of View answers nearly every question the new cat owner could have and gives the experienced cat owner a look at life from the other side of the...»

«THE SIGNIFIER POINTING AT THE MOON THE SIGNIFIER POINTING AT THE MOON Psychoanalysis and Zen Buddhism Raul Moncayo Masao Abe, Zen and Western Thought, 1999, University of Hawaii Press, reproduced with permission of Palgrave Macmillan. Ekaku Hakuin (1685–1788). Daruma, 18th century. Hanging scroll, ink on paper © 2012 Museum Associates/LACMA. Licensed by Art Resource, NY. First published in 2012 by Karnac Books Ltd 118 Finchley Road London NW3 5HT Copyright © 2012 by Raul Moncayo The right...»

«Begin Reading Table of Contents Copyright Page In accordance with the U.S. Copyright Act of 1976, the scanning, uploading, and electronic sharing of any part of this book without the permission of the publisher constitute unlawful piracy and theft of the author’s intellectual property. If you would like to use material from the book (other than for review purposes), prior written permission must be obtained by contacting the publisher at permissions@hbgusa.com. Thank you for your support of...»

«29 Hacker Perspectives on Privacy “It Doesn’t Have to Be This Way”: Hacker Perspectives on Privacy Kevin Steinmetz & Jurg Gerber* We face a lot of troubling times ahead with regards to surveillance. Most of the power, for the moment at least, remains in our hands and in our minds, should we choose to use them. It is our acceptance of the elements of a surveillance state which will give it the most strength and solidify its presence for future generations. It doesn’t have to be this way....»

«TECHNISCHE UNIVERSITÄT MÜNCHEN Lehrstuhl für Aquatische Systembiologie Bioindication tools for measuring the success of stream restoration Joachim Wolfgang Pander Vollständiger Abdruck der von der Fakultät Wissenschaftszentrum Weihenstephan für Ernährung, Landnutzung und Umwelt der Technischen Universität München zur Erlangung des akademischen Grades eines Doktors der Naturwissenschaften genehmigten Dissertation. Vorsitzender: Univ.-Prof. Dr. A. Melzer Prüfer der Dissertation: 1....»

«From Soulard’s Notebooks Assistant Editor: Kassandra Soulard Poetry by Ric Amante 1 Notes from New England [Commentary] by Raymond Soulard, Jr. 9 Poetry by Judih Haggai 27 Poetry by Joe Ciccone 32 Paelo Redemption [Travel Journal] by Charlie Beyer 35 Poetry by Joe Coleman 45 Baby Steps Up into the Sky [Travel Journal] by Nathan D. Horowitz 55 Many Musics [Poetry] by Raymond Soulard, Jr. 59 Psychedelics and Lucid Dreaming: Doorways in the Mind [Essay] by A.S. Kay 91 Poetry by Tom Sheehan 99...»

«*** Free EBOOK*** Your Journey of Personal Transformation Begins HERE: An Introduction to Understanding Your Human Design Chart by Evelyn Levenson, MBA Certified Human Design Specialist Your Success By Design Coach Your Journey Begins HERE www.HumanDesignforSuccess.com This Report is for your personal use. You MAY SHARE it, in its entirety with all author and contact information. However, You May NOT Trade, Sell or Resell This Intellectual Property in Any Way. You may not reprint any part of it...»

«Fools Die Mario Puzo Book I Chapter 1 “Listen to me. I will tell you the truth about a man’s life. I will tell you the truth about his love for women. That he never hates them. Already you think I’m on the wrong track. Stay with me. Really—I’m a master of magic. “Do you believe a man can truly love a woman and constantly betray her? Never mind physically, but betray her in his mind, in the very ‘poetry of his soul.’ Well, it’s not easy, but men do it all the time. “Do you...»

«Diffusion: the UCLan Journal of Undergraduate Research Volume 4 Issue 2 (December 2011) The role of religion in social transformation within Poland since1970 Paul Ryder (Religion, Culture and Society) The political and social structure of the state At the close of the Second World War in 1945, Poland was consumed by the totalitarian, communist regime of Joseph Stalin’s Soviet Union. The ideology of Stalinism was heavily based on the work of Karl Marx and what later became known as Marxism....»

«CANCER SUPPORT SERVICES in Ireland DIRECTORY Leinster Irish Cancer Society DIRECTORY OF SERVICES Welcome Welcome to the Fourth Edition of the Directory of Cancer Support Services Affiliated to the Irish Cancer Society. More people are getting cancer in Ireland. It is estimated that one in three of us will be diagnosed with the disease during our lifetime. Cancer is however increasingly viewed as a condition from which people survive. Evidence shows that when people experiencing cancer receive...»

«Begin Reading Table of Contents Copyright Page In accordance with the U.S. Copyright Act of 1976, the scanning, uploading, and electronic sharing of any part of this book without the permission of the publisher constitute unlawful piracy and theft of the author’s intellectual property. If you would like to use material from the book (other than for review purposes), prior written permission must be obtained by contacting the publisher at permissions@hbgusa.com. Thank you for your support of...»

<<  HOME   |    CONTACTS
2016 www.dissertation.xlibx.info - Dissertations, online materials

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.