Jan Puzicha and Joachim Buhmann. Multiscale annealing for real-time unsupervised texture segmentation. In Proceedings of the Int. Conf. Comp. Vision. IEEE, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of an initial texture model [3] 5] This is usually a computational expensive task requiring simulated annealing or similar iterative schemes. The extension of the MRF model to a multiscale random field (MSRF) that is hierarchically ordered, reduces the computation time significantly [2] 1] [8] [7] 6] 9] As a disadvantage some MSRF models are either not exactly tractable or produce rather blocky regions. In this paper we present a model that avoids both of these drawbacks. We let each random variable of the MSRF choose its parent during segmentation, so that the preferred locations ....

Jan Puzicha and Joachim Buhmann. Multiscale annealing for real-time unsupervised texture segmentation. In Proceedings of the Int. Conf. Comp. Vision. IEEE, 1998.

....Markov random fields with smooth spatial behavior [3] They are usually computationally expensive requiring simulated annealing or similar iterative schemes. The extension to multi scale Markov random fields (MSRF) which are hierarchically ordered, reduces the computing time significantly [5] [6] [7] As a disadvantage some MSRF models are either not exactly tractable or produce rather blocky regions. Here we model within a Bayesian framework the pixel labels as a MSRF. As an essential extension of the approach of Bouman and Shapiro [5] we optimize dynamically the structure of the ....

J. Puzicha and J. Buhmann, "Multiscale annealing for real-time unsupervised texture segmentation, " in Proceedings of the Int. Conf. Comp. Vision, IEEE, 1998.

....over a signicant area. However, most real images regions do not present stationary features. Also, unmeaningful small sized regions related to stains, noise or punctual information may appear. Consequently, methods relying on a priori knowledge of the number of textures in a given image [8] 11] [9] often fail, because if any unexpected texture region appears, like the ones related to shadows or boundaries, a wrong fusion of two non related regions is forced. Unsupervised segmentation [4] 6] 2] does not rely on such a knowledge, but it is slower because it requires a computationally ....

....but it is slower because it requires a computationally expensive additional stage to calculate the correct number of regions in the image [4] A common choice to speed up segmentation processes is to use multiresolution structures. Most multiresolution methods work in a coarse to ne way [8] [9] [2] 10] Their main problem is that they typically adopt the same cost functions at all scales [8] 2] Unfortunately, cost functions tend to work ne either at high or at low resolution. Our eorts have been focused on implementing an unsupervised texture based segmentation algorithm to achieve ....

[Article contains additional citation context not shown here]

J. Puzicha and J.M. Buhman , "Multiscale anneal ing for realtime unsupervised texture segmentation", Technical Report IAITR -97-4, Rhein ische Friedrich-Wilhelms-Universitt, 1997

.... Some exemplary application areas of dyadic data are: ffl Computer vision, in particular in the context of image segmentation, where X corresponds to image locations, Y to discretized or categorical feature values, and a dyad denotes the occurrence of a feature at a particular image location [HPB98] ffl Text based information retrieval, where X corresponds to a document collection, Y to the vocabulary, and a dyad represents the occurrence of a token in the content of a document [HPJ99] ffl Computational linguistics in the corpus based statistical analysis of word co occurences which has ....

....OEg QfC(x) cjS; OEgQfD(y) djS; OEg ; 29) and to utilize this approximation for the parameter re estimation in the M step. Here, Q is an approximating probability distribution which is chosen in order to minimize the KL divergence to the true posterior distribution (cf. NH98, JGJS98, HPB98] The approximate E step equations are given by (cf. Appendix) QfC(x) cjS; OEg P (c) exp X y n(x; y) X d QfD(y) djS; OEg log OE(c; d) # ; 30) QfD(y) djS; OEg P (d) exp X x n(x; y) X c QfC(x) cjS; OEg log OE(c; d) # : 31) Notice that the mean field conditions form a highly ....

[Article contains additional citation context not shown here]

Jan Puzicha and Joachim Buhmann. Multi--scale annealing for real--time unsupervised texture segmentation. In Proceedings of the International Conference on Computer Vision (ICCV'98), pages 267--273, 1998.

....color and texture features. Earlier work has used EM and or the Minimum Description Length (MDL) principle to perform segmentation based on motion [3, 43] or scaled intensities [44] but EM has not previously been used on joint color and texture. Related approaches such as deterministic annealing [33] and classical clustering [22] have been applied to texture segmentation without color. Panjwani and Healey [30] have performed segmentation using a Markov random field color texture model. 2 Feature extraction Creating the Blobworld representation of an image involves three steps (see Figure 1) ....

J. Puzichaand J. M. Buhmann. Multiscale annealing for realtime unsupervised texture segmentation. In Proc. Int. Conf. Comp. Vis., 1998.

.... distributional clustering provides a generative statistical model that can be utilized in subsequent processing steps such as boundary localization [10] Distributional clustering also offers computational advantages, because it can be implemented efficiently by multiscale optimization techniques [8]. Since feature histograms are processed directly, it avoids time consuming stages of data extraction (e.g. pairwise comparisons in PDC) which is crucial in real time applications like autonomous robotics. 2 Mixture Models for Histogram Data Model Specification To stress the generality of the ....

....taken from aerial images (ground truth unavailable) hood by maximizing over a suitable nested sequence of subspaces in a coarse to fine manner, where each subspace has a greatly reduced number of class assignment variables. This strategy is formalized by the concept of multiscale optimization [3, 8] which in essence leads to cost functions redefined on a coarse version of the original image. In contrast to most multi resolution optimization schemes, the original cost function is optimized at all grids, only the configuration space is reduced by variable tying. We first sketch the general ....

[Article contains additional citation context not shown here]

J. Puzicha and J. Buhmann. Multi--scale annealing for real--time unsupervised texture segmentation. In Proc. International Conference on Computer Vision (ICCV'98), pages 267--273, 1998.

.... the texture segmentation context, which we refer to as pairwise dissimilarity clustering (PDC) Although these methods are directly applicable to proximity data, they are only tractable in image segmentation problems if they avoid the computation of dissimilarities for all possible pairs of sites [9]. The major contribution of this paper is a general approach to the problem of grouping feature distributions, extending a technique known as distributional clustering in statistical language modeling [8] In contrast to methods based on feature vectors and pairwise dissimilarities this approach ....

....measure, but exclusively relies on the feature occurrence statistics. Another important consideration for a clustering approach to image segmentation are real time constraints. Given the respective data (vectors, histograms or proximities) all algorithms require only a few seconds for optimization [9]. While vector based methods suffer from inferior quality, it is the data extraction process of PDC which is prohibitive for real time applications like autonomous robotics. Using the histogram data directly avoids the necessity for pairwise comparisons altogether while achieving segmentations ....

[Article contains additional citation context not shown here]

J. Puzicha and J. Buhmann. Multiscale annealing for real--time unsupervised texture segmentation. Technical Report IAI--97--4, Institut fur Informatik III (a short version appeared in: Proc. ICCV'98, pp. 267--273), 1997.

....a model validation criterion by optimizing over different K [2, 4] To unify the representation of a segmentation we introduce Boolean assignment variables M i 2 f0; 1g denoting whether image site i is labeled as texture class . All assignments 2 An extended abstract of this paper appeared in [35]. J. Puzicha, J.M. Buhmann: Real Time Texture Segmentation 4 are summarized in terms of a Boolean assignment matrix M 2 MN;K , where MN;K = M 2 f0; 1g N ThetaK : K X =1 M i = 1 ) 1) In optimization approaches, a real valued objective function H(M 2 MN;K ) is designed which ....

J. Puzicha and J. Buhmann, "Multi--scale annealing for real--time unsupervised texture segmentation, " in Proceedings of the International Conference on Computer Vision (ICCV'98), pp. 267--273, 1998.

....smoothing by optimizing over a probabilistic state space. DA is applicable to non linear cost functions and yields efficient algorithms which possess a favorable scaling behavior in terms of computational complexity. DA has empirically shown to compute optimal or near optimal solutions [6, 4, 7], which makes it a promising optimization heuristic for large problem instances. Global optimality, however, has not been established yet even for carefully annealing. We present a rigorous mathematical analysis of DA, which culminates in establishing a close connection to homotopy and ....

....exploited to combine the basic corrector step from Theorem 2 with a multi grid inspired course to fine technique to achieve highly efficient optimization methods capable of the real time demands in autonomous robotic applications. The resulting algorithm is refered to as multiscale annealing [7]. To demonstrate the capability of the proposed algorithms an exemplary result on real world data with a problem size of 12288 Boolean variables is given in Fig. 1 (1:21s CPU time on a SUN UltraSparc) A second example with a problem size of 16382 Boolean variables is given in Fig. 2, where the ....

[Article contains additional citation context not shown here]

J. Puzicha and J. Buhmann. Multiscale annealing for real--time unsupervised texture segmentation. Technical Report IAI--97--4, Institut fur Informatik III, Universitat Bonn, 1997.

No context found.

J. Puzicha and J. Buhmann. Multiscale annealing for real--time unsupervised texture segmentation. Technical Report IAI--97--4, Institut fur Informatik, Universitat Bonn (a short version appeared in: Proc. ICCV'98, pp. 267--273), 1997.

....convergence. In case that a constraint is violated after performing an overrelaxed M step, the parameter set is projected back on the admissible parameter space. For an overview on more elaborated acceleration methods for EM we refer to [36] 5. 4 Multiscale Optimization Multiscale optimization [21, 46] is an approach for accelerating clustering algorithms whenever a topological structure exists on the object space(s) In image segmentation, for example, it is a natural assumption that adjacent image sites belong with high probability to the same cluster or image segment. This fact can be ....

J. Puzicha and J. Buhmann. Multiscale annealing for real--time unsupervised texture segmentation. Technical Report IAI--97--4, Institut fur Informatik III, 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC