Polygonal Approximation of Point Sets
|
|
- Sydney Morton
- 5 years ago
- Views:
Transcription
1 Poygona Approximation of Point Sets Longin Jan Latecki 1, Rof Lakaemper 1, and Marc Sobe 2 1 CIS Dept., Tempe University, Phiadephia, PA 19122, USA, atecki@tempe.edu, akamper@tempe.edu 2 Statistics Dept., Tempe University, Phiadephia, PA 19122, USA, marc.sobe@tempe.edu Abstract. Our domain of interest is poygona (and poyhedra) approximation of point sets. Neither the order of data points nor the number of needed ine segments (surface patches) are known. In particuar, point sets can be obtained by aser range scanner mounted on a moving robot or given as edge pixes/voxes in digita images. Poygona approximation of edge pixes can aso be interpreted as grouping of edge pixes to parts of object contours. The presented approach is described in the statistica framework of Expectation Maximization (EM) and in cognitivey motivated geometric framework. We use oca support estimation motivated by human visua perception to evauate support in data points of EM components after each EM step. Consequenty, we are abe to recognize a ocay optima soution that is not gobay optima, and modify the number of mode components and their parameters. We wi show experimentay that the proposed approach has much stronger goba convergence properties than the EM approach. In particuar, the proposed approach is abe to converge to a gobay optima soution independent of the initia number of mode components and their initia parameters. 1 Introduction Expectation Maximization (EM) is a very popuar and powerfu method that aows simutaneous estimation of mode parameters and assignment of data points to components of the mode. However, EM produces an optima soution ony if the number of mode components is we estimated and the initia vaues of mode parameters are cose to the goba optimum. If this is not the case EM is ony guaranteed to produce a ocay optima soution. This is iustrated in Fig. 1, where (a) shows data points and the initia configuration of two straight ine segments. The number of mode components (2 ine segments) is correcty initiaized, but their position is not sufficienty cose to the goba optimum. Fig. 1(b) shows the fina, ocay optima, resut obtained by the cassica EM agorithm. Fig. 1(c) shows the gobay optima approximation obtained by the proposed method on the same input. Due to the oca optimum probem, a correct estimation of the number of components and the initia parameters of a statistica mode is crucia in a EM appications, and therefore, beongs to one of the most chaenging probems in statistica reasoning. The proposed approach provides a soution to the probem
2 (a) Ange 9.; Standard Deviation. segments: 2 subsegeng.1 (b) (c) Fig. 1. (a) shows the data points and the initia position of mode ines. (b) shows the optima approximation of the data points obtained by EM. (c) shows the optima approximation resut obtained by the proposed method. of oca optimum in EM that is based on cognitivey motivated oca support evauation of EM components. The exampe shown in Fig. 2 motivates the proposed approach. It is obvious to humans that the approximation in (c) of the underying data points is significanty better then the approximation in (a). Observe the ack of oca support in the data points of the midde part of the ine in (a). This observation is the key argument for the proposed extension of EM. We wi evauate the support in data points of each EM component, and remove parts of components with insufficient support. The existing approaches to determine the optima number of EM components, of which AIC and BIC (Bayesian Information Criterion), which is equivaent to MDL (Minimum Description Length) [1], are most known, do not base their decision on the oca support in data points of the mode components. They assume ony a fix cost per each mode parameter. In particuar, this means that a mode component with high data support (i.e., positioned in a data region with high point density) costs the same as a component with ow data support (i.e., positioned in a regions with ow density of data points) athough it is intuitivey cear that a component with ow data support is far ess reevant than a component with arge data support. AIC, BIC, and MDL require separate EM runs unti convergence for a possibe number of mode components, each run composed of severa EM iterations,
3 69 iteration 2, no spitting 69 iteration 2, no spitting (a) (b) iteration (c) Fig. 2. It is obvious to us that the approximation in (c) of the underying data points is significanty better then the approximation in (a). (a) shows the best possibe approximation of the data points obtained by EM. (b) iustrates the ine spit (LS) based on subsegment remova. The removed subsegments are marked with crosses. (c) shows the fina approximation resut obtained by EM after the spit. which may even be in the order of severa thousands. For AIC, BIC, and MDL to be successfu, it is impicity assumed that EM converges to goba optimum in each run. However, as iustrated in Fig. 1(b) this is not aways the case, since even with a correct number of components EM may get stuck in oca minimum. Therefore, the correct number of two mode components woud not be seected by AIC, BIC, or MDL in our exampe. By ocay evauating the support in the data points of the two ines in Fig. 1(b), we can ceary determine that they form a bad approximation of the data points. By removing most of their parts, and retaining ony sma parts around the data points, we create a better input for the EM agorithm. This finay eads to a gobay optima approximation in Fig. 1(c). The proposed approach provides a soution to the probem of oca optimum in EM by adding two new steps that are we integrated with the standard E and M steps of EM. The two new steps are geometricay motivated and can be interpreted as spit and merge steps in the context of ine fitting. However, the proposed extension of EM is not restricted to any particuar shape of mode components. In the first new step, the spit step, the mode components obtained by a previous EM iteration are examined for support of the data points. The main idea (iustrated by the above exampe) is that higher point density around a mode component (ine segment in our appication) indicates a presence of a inear structure in the data points around the segment. Parts of the segment that do not have sufficient support are removed. This may ead to segment remova
4 but generay eads to a spit of the segment into severa subsegments. The second new step is merging simiar mode components. It prevents generating statistica modes that overfit the data, i.e., fit noise in the data. This step requires a simiarity measure of statistica mode components. The merging step can be interpreted as perceptua grouping that dates back to the first resuts of Gestat psychoogy in the beginning of 2th century [2]. We wi show that integrating the spit and merge operations in the EM framework eads to a gobay optima soution. In our experiments, we were abe to obtain a gobay optima soution after just a few iterations (between and 3). Two exampe appications of our approach are outined in Fig. 3. (a) shows an origina input toy image. (b) shows the edges obtained by Canny edge detector with a substantia amount of added noise, and the initia mode for our agorithm. It consists of ony two ine segments. (c) shows an intermediate step of our agorithm. The fina poygona approximation obtained after 27 iterations is shown in (d). (e) shows an image obtained by samping 3 ground truth segments (1 points) with a substantia amount of noise (2 points). (f) shows the initia mode segments for our agorithm. We present the resuts of our agorithm after 8 in (g) and 19 iterations in (h). An overview of techniques for poygona approximations of curves (when the order of data points is known), which have been studied at east since eary seventies in computer vision, can be found in [3]. To some popuar greed poygona approximation methods in digita images beong [4] and [, 6]. An overview of approaches to obtain poygona maps from aser range data can be found in [7, 8]. We do not make any assumptions about the order of data points and extent of noise. The proposed method avoids the probem of a ocay optima soution and produces stabe approximations not ony to straight but aso to curved ines. Moreover, the fina number of fitted ine segments depends on extent of noise. This means that the number of mode components is adjusted to achieve the best possibe approximation accuracy as the function of noise extent. In order to show that the geometric and cognitivey motivated spit and merge steps can be incorporated into a statistica formaism, we introduce in Section 2 a new target function to be estimated in the EM framework, and reformuate the E and M steps in Section 3. Then we introduce statistica tests for the proposed spit and merge steps in Section 4. Finay in Section, we describe the geometric parts of the spit and merge steps. 2 Optimizing Kuback-Leiber Divergence Our goa is to approximate the ground-truth density q(x) with a member p Θ (x) of a parametric famiy {p Θ (x) : Θ S} of densities. We use Kuback-Leiber divergence (KLD) to measure dissimiarity between the ground-truth and parametric famiy of densities. By definition, the KLD between the ground truth q(x)
5 iteration 1, after (merge) & EM iteration 2, after (merge) & EM (a) (b) (c) iteration 27, after spit, before EM 2 iteration 1, after (merge) & EM (d) (e) (f) iteration 4, after spit & EM iteration 19, after spit & EM (g) (h) Fig. 3. (a) An origina input image. (b) The edges obtained by Canny edge detector, and two initia ine segments. (c) We see the poygona approximation of the edge pixes obtained after (d) The fina poygona approximation obtained after 27 iterations is shown in (e)-(g) Iustrate our approach on simuated data generated by 3 ground truth segments with ony 1 signa and 2 noise points. and the density, p Θ (x) is: D(q(x) p Θ (x)) = og q(x) p Θ (x) q(x)dx = og q(x)q(x)dx og p Θ (x)q(x)dx (1) Observe that KLD is abe to determine the optima number of mode components of p Θ. This is due to the fact that KLD D(q p Θ ), viewed as a functiona on the space { p Θ } of Gaussian mixtures, is convex and hence has a unique minimum. It can be easiy derived that the parameters Θ minimizing (1) are given by Θ = argmax Θ { og p Θ (x)q(x)dx } (2)
6 We obtain the cassica maximum ikeihood estimator by appying the MC (Monte Caro) integra estimator to (2) under the assumption that the observations x 1,..., x n are i.i.d. (independenty and identicay distributed) sampe points seected from the distribution q(x). Θ = argmax Θ og p Θ (x i ) (3) However, as we derive beow (equation (6)), if some proportion of the observations x 1,..., x n are noisy, a more accurate estimator of Θ in (2) is given by: Θ = argmax θ og p θ (x i )sdd(x i ), (4) i where sdd is caed the smoothed data density and is defined in (7) beow. Equation (4) is the basis of the proposed approach. We demonstrate theoreticay and experimentay that maximization in (4) yieds substantiay better resuts than the cassica EM maximization in (3). To demonstrate the significance of (4), we consider the probem of estimating the optima number of mode components by minimizing the KLD D(q(x) p Θ (x)) in Θ. The parametric famiy { p Θ (x) }, being a famiy of Gaussian mixture distributions, is convex. It foows that there is a unique member (x) of the Gaussian mixture famiy p Θ with minimum KLD from q. This minimizing mixture must have the correct number of mode components. However, it is we known that (3) cannot be used to estimate the correct number of mode components, since (3) increases when the number of mode components increases. In contrast, we are abe to determine the correct number of mode components when using (4) to estimate the KLD, D(q(x) p Θ (x)). Thus, the modified EM agorithm that maximizes (4) is not ony abe to estimate mode parameters but aso the right number of mode components. One of the key steps in the derivation of (4) is the Monte Caro (MC) estimate of the integra given by the right hand side of equation (1). Let x 1,..., x n be i.i.d. sampe points drown from the probabiity density function (pdf) q(x). Then we can approximate the integra of a continuous function f by its MC estimate: f(x)q(x)dx 1 f(x i ) () n In the usua approach to inference, it is a commony accepted assumption that sampe data points x 1,..., x n are distributed according to the (estimated) density q(x). This assumption is the key to insuring that maximum ikeihood estimators are appropriate for purposes of estimating parameters of interest. However, in a rea appications, the sampe data points are corrupted by a certain amount of noise. Usuay the proportion of noisy points does not decrease when the number of sampe points is increased. Due to the noise, the foowing equation provides a substantiay better estimate f(x)q(x)dx f(x i )sdd(x i ). (6) i i i
7 Finay equation (4) ceary foows from (6) and (2). The smoothed data density sdd is defined as sdd(x) K( d(x, x i) ) = 1 h nh G(d(x, x i ),, h), (7) where proportionaity refers to the fact that sdd(x i ) = 1, d(x, y) is the Eucidean distance, and G(d(x, y),, h) is a Gaussian with mean zero and the standard deviation (std) h. An intuitive motivation for sdd is as foows: If a given data point x j is samped from the true distribution q(x), then x j ies in a dense region of the observed sampe points and consequenty sdd(x j ) is arge. If a given data point x j is samped from the noise distribution, then x j is ikey to ie in a sparse region of the sampe space, and consequenty sdd(x j ) is sma. To estimate the bandwidth parameter h, we can draw from a arge iterature on nonparametric density estimation [9, 1]. As we show in the presented experimenta resuts, an accurate bandwidth estimation in not crucia in our approach. 3 E and M Steps We introduce atent variabes z 1,..., z n which serve to propery abe the respective data points x 1,..., x n. It is assumed that the pairs (x i, z i ) for i = 1,..., n are i.i.d. with common (unknown) joint (ground truth) density, q(x, z) = q(x)q(z x); q(x) is the margina x-density and q(z x) is the conditiona density of the abe z given x. In this new framework, the KLD between the joint density q(x, z) and a parametric counterpart density p Θ (x, z) is D(q(x, z) p Θ (x, z)) = D(q(x)q(z x) p Θ (x)p Θ (z x)) { [ ] [ ]} q(x) q(z x) = og + og q(x)q(z x)dzdx x z p Θ (x) p Θ (z x) [ ] [ ] q(x) q(z x) = og q(x)dx + q(x) og q(z x)dz (8) p Θ (x) p Θ (z x) x x We are now ready to introduce the expectation (E) and maximization (M) steps. Both steps aim at minimizing the same target function (8) in our framework. The expectation step yieds the standard EM formua; considerations discussed above ead to a different soution for the maximization step. Expectation Step: For a fixed set of parameters Θ, we want to find a conditiona density q(z x) that minimizes D(q(x, z) p Θ (x, z)). Since KLD is aways nonnegative, and the second summand in (8) is minimized for q(z x) = p Θ (z x) (in which case it is equa to zero), we obtain from (8) that q(z x) = p Θ (z x) minimizes D(q(x, z) p Θ (x, z)). z
8 In particuar, for given sampe points x 1,..., x n, we obtain q(z i = x i ) = p Θ (z i = x i ) = p(z i = x i, Θ) (9) = p(x i z i =, Θ)p(z i = Θ) p(x i Θ) (1) p(x i z i =, Θ)p(z i = Θ) = k j=1 p(x i z i = j, Θ)p(z i = j Θ) = p(x i z i =, Θ)π k j=1 p(x, i z i = j, Θ)π j (11) where π = p(z i = Θ) and π j = p(z i = j Θ) are the prior probabiities of component abes and j correspondingy. Maximization Step: For the fixed margina distribution q(z x) = p Θ (z x), we want to find a set of parameters Θ that maximizes (8). Substituting q(z x) = p Θ (z x) in (8), we obtain D(q(x, z) p Θ (x, z)) = og( q(x) p Θ (x) )q(x)dx = D(q(x) p Θ(x)) (12) Thus, minimizing D(q(x, z) p Θ (x, z)) in Θ is equivaent to minimizing D(q(x) p Θ (x)) in Θ. Using the estimate derived in equation (4), minimizing (12) in Θ is equivaent (in the MC setting discussed above) to maximizing the weighted margina density W M(Θ) = sdd(x i ) og p Θ (x i ) = sdd(x i ) og p(x i Θ) = = k sdd(x i ) og[ p(x i z i =, Θ)p(z i = Θ)] =1 k sdd(x i ) og[ p(x i z i =, Θ)π ] (13) =1 where π = p(z i = Θ) are the prior probabiities of component abes = 1,..., k. Now we expicity use the incrementa update steps of the EM framework. Using the prior probabiities of component abes π (t) = p(z i = Θ (t) ) obtained at stage t for = 1,..., k, we obtain from (13) that an update of W M(Θ) is estimated by maximizing W M(Θ; Θ (t) ) = k sdd(x i ) og[ p(x i z i =, Θ)π (t) ] (14) in Θ with Θ (t) denoting the vaue of Θ computed at stage t of the agorithm. The crucia difference between this and the standard EM update is that our target function is weighted with terms sdd(x i ). We note that the known convergence proofs for the EM agorithm appy in our framework, since adding the weights sdd(x i ) in (14) does not infuence the convergence. =1
9 4 Spit and Merge The proposed spit and merge steps adjust the number of mode components by performing compnent spit and merge steps ony if they increase the vaue of our target function (14). Our framework is very genera in that it aows many possibe seections of the candidate components for the spit and merge steps. We present specific seection methods of the candidate components in Section. They are based on a Maximum A Posteriori principe. In the foowing formuas, we assume that the candidate components are given. Spit: Assume that we are given two candidate mode components 1, 2 ; we consider repacing the mode component with components 1, 2. Since our goa is maximizing QM(Θ; Θ (t) ) in formua (14), we simpy need to check whether repacing with 1, 2 increases W M, where j {1,..., k}: < W M(Θ; Θ (t) ) = sdd(x i ) og[ j sdd(x i ) og[ j + p(x i z i = 1, Θ)π (t) 1 p(x i z i =, Θ)π (t) p(x i z i = j, Θ)π (t) j ] + p(x i z i = 2, Θ)π (t) 2 ] (1) We ony need to perform oca computation to perform this test, i.e., we ony need to compute the corresponding probabiities for the candidate components 1, 2, subject to the condition that π (t) = π (t) 1 + π (t) 2. The parameters are estimated foowing the sparse EM step in Nea and Hinton [11], (see equation (1)). In accordance with the resuts of [11] this oca computation guarantees that the target function increases after each iteration (if (1) hods). Convergence is aso guaranteed in this way. Merge: Given a candidate component, we merge two existing mode components 1, 2 to if for j {1,..., k} > W M(Θ; Θ (t) ) = sdd(x i ) og[ j sdd(x i ) og[ j + p(x i z i = 1, Θ)π (t) 1 p(x i z i =, Θ)π (t) p(x i z i = j, Θ)π (t) j ] + p(x i z i = 2, Θ)π (t) 2 ] (16) Again we ony need to perform oca computations to perform this test. For merge, we ony need to compute the corresponding probabiities for the candidate component, subject to the same constraint π (t) = π (t) 1 + π (t) 2. If (16) hods and we repace 1, 2 with, the convergence of our agorithm foows from the resuts of [11]. We note that the proposed spit and merge steps do not work in the cassica EM framework. To see this, consider sdd(x i ) = 1 for a the data points
10 (i = 1,..., n). The merge inequaity (16) is not satisfied even if the ground truth mode is assumed to be a singe component, since mutipe components can better fit the data, and consequenty have a arger og ikeihood vaue. Anaogousy, if the spit inequaity (1) hods for a reasonabe seection of candidate component modes, the cassica EM framework incorrecty spits ground truth components. Thus, a mixture mode of arger number of components is aways prefered in the cassica EM framework. In the proposed framework, sdd represents an estimated density of the data points. Consequenty, in the proposed spit and merge steps, the divergence of parametric components, 1, 2 from the ground truth is evauated with respect to this nonparametric density. Line Segments as Components We present specific detais concerning our use of ine segments as EM mode components in the appications presented beow. We stress that this section appies aso to hyper panes in any dimensions, but the presentation is given in terms of ine segments for purposes of simpification. The proposed approach requires a minor extension of EM ine fitting to work with ine segments, which we wi ca Expectation Maximization Segment Fitting (EMSF). The difference between EMSF and EM ine fitting is that our mode components are ine segments (rather than ines). The input, for our mode, is a set of ine segments and a set of data points. As with EM the proposed EMSF is composed of two steps: (1) E-step The EM probabiities are computed based on the distances of points to ine segments instead of the distances of points to ines. (2) M-step Given the probabiities computed in the E-step, the new positions of the ines are computed by minimizing squared regression error weighted with these probabiities. As in the case of EM ine fitting, the output of the M-step is a new set of ines (not ine segments). Since we need ine segments as input to the E-step, we trim ines to ine segment based on their support in the sampe data. This is done by the spit process described in Section.2. Now we describe the specific detais reated to ine segments for steps (1) and (2). In order to derive the soution of (14) for EM mode components being ine segments, we introduce so caed EM weights. In the cassica EM, the weight = p(z i = x i, Θ (t) ) represents the probabiity that x i corresponds to segment s for = 1,..., k. We use the notation θ for the parameters of the ine segment s itsef. In our framework w (t) i w (t) i sdd (t) (x i ) p(z i = x i, Θ (t) ), (17) and the weights are normaized so that k =1 w(t) i = 1 for each i. After the E- step associated with the t th iteration is accompished, we obtain a new matrix ). Intuitivey, each row i = 1,..., n of this matrix corresponds to weighted (w (t) i
11 probabiities that the data point x i is associated with the corresponding ine segments; each coumn = 1,..., k can be viewed as weights representing the infuence of each point on the computation of new ine positions in the M-step. Beow, we use the notation x i = (x ix, x iy ) with (i = 1,..., n) for the coordinates of the observed data points, and ( x, ȳ) for the coordinate averages. The ine L, constructed beow, is constructed to go through the point ( x, ȳ). To obtain the soution of (14), we perform an orthogona regression weighted with the matrix (w i ). The soution is given as the norma vector to ine L, which is the vector corresponding to the smaest eigenvaue of the matrix M defined as [ n w i(x ix x) 2 n w ] i(x ix x)(x iy ȳ) n w i(x ix x)(x iy ȳ) n w i(x iy ȳ) 2 (18) Finay the parameters θ (t+1) are given as parameters of the ine segment s (t+1) obtained by trimming the ine L to the data points. We are now ready to introduce particuar reaization of spit and merge for EM mode components being ine segments. The proposed spit and merge EM segment fitting (SMEMSF) agorithm iterates the foowing three steps (1) EMSF (2) Spit (3) Merge Spit step is presented in detai in Section.2 whie Merge step is described in Section.1. Spit evauates the support in the data points of ines obtained by EMSF and removes the parts that are weaky supported. Since we have a finite set of data points, this has the effect of trimming the ines to ine segments. Finay the merge step merges simiar ine segments. Thus, spit and merge steps adjust the number of mode components to better fit the data..1 Merging If inequaity (16) hods, we merge two mode components represented by parameters 1, 2 into one mode componet given by parameter. Whie components 1, 2 are present at step t (they are ine segments s 1, s 2 ), we did not yet specify hot to compute the candidate component. Now we describe a particuar method to generate a candidate component in the particuar case in which the mode components are ine segments. We stress that other methods are possibe and that inequaity (16) appies to them too. A support set S(s j ) for a given ine segment s j (mode component ) is defined as set of points whose probabiity of supporting segment s j is the argest, i.e., S(s j ) = {x i : w ij = max(w i1,..., w ik )}. This maps each data point to a unique segment using the Maximum A Posteriori principe. Given two ine segments s 1, s 2, the merged segment s is obtained by trimming the straight ine obtained by regression on data points in S(s 1 ) S(s 2 ). Trimming is performed by ine spit described in Section.2.
12 .2 Line spit (LS) A cassica case of EM oca optimum probem is iustrated in Fig. 2(a), where the ine segment is in a ocay optima position. Ceary, the probem here is that we have a mode consisting of one ine ony, whie two ine segments are needed. Fig. 2(b) iustrates a spit operation described in this section. It is based on remova of subsegments that do not have sufficient support in the data points. As the resut we obtain two ine segments. Finay, Fig. 2(c) shows the gobay optima approximation of the data points obtained by EM appied to the two segments. The main idea is that higher point density aong a segment indicates a presence of a inear structure in the data points around the segment. Each ine or ine segment is examined on having sufficient support in data points measured as point density around it. Ony parts of segments that have sufficient support of the data points remain. This eads to spit of existing ines or segments aowing us to adjust the number of the ine segments (i.e., the number of EM mode components) to better fit the input data points. In [12] the number of points in discretey enumerated rectanguar strips (e.g., neighborhoods of a possibe segments with endpoints in some finite set) is counted. Then signa strips are seected based on the ratio of the number of points to the area. Then strips are grouped to poyines based on proximity of their endpoints and their anges. The main difference of our approach to the approach in [12] is that we do not seect the signa segments, but evauate the existing structures seected by EM. This makes our computation more efficient, since we do not need to numerate a possibe strips, and more accurate, since the ine segments are optimay fitted to the data points in our approach. Line Spit (LS) is composed of the foowing steps: (2.1) Subsegment support computation. (2.2) Remova of subsegments with insufficient support that satisfy inequaity (1). The input to LS are segments s 1,..., s k obtained by cipping the ines 1,..., k created in EMSF to the image rectange. We divide each segment s j {s 1,..., s k } into subsegments of a predefined ength 2r, i.e., s j = I j 1... Ij, so that two consecutive subsegments overap in their common endpoint, where is the number of subsegments. (For simpicity we assume that the ength of s j is exacty mutipe of 2r.) For each subsegment I j k, we define its support as the number of data points in the square S(I j k ) whose two sides are parae to subsegment I j k and whose center is contained in I j k, i.e., support(i j k ) = #({x i} S(I j k )). A few such squares are iustrated in Fig. 2(b). In each iteration a support threshod C is computed from the statistics of support(i j k ) vaues over a subsegments of a ine segments. Finay subsegments I j k with support(i j k ) C are removed. The subsegments to be removed are marked with crosses in Fig. 2(b). New segments are created as connected
13 components of remaining subsegments of segment sj. If inequaity (1) hods, then the origina input segment sj is removed, and the newy created segments are added to the ist of origina segments for the next iteration of EMSF. If a its subsegments are removed, then a given segment is removed. 6 Appications Two exampes of approximations of point sets in digita images are iustrated in Fig. 3. An exampe appication of our approach in robot mapping is outined in Fig. 4. (a) shows an origina data set of aser range scan points aigned with the agorithm presented in [13]. The origina set is composed of 39 scans, each with 361 points. Thus, the origina input map is composed of 142,9 points. We initiaize our agorithm with 192 segments, the grid segments, as mode components. (b) shows the output with 96 segments after the first iteration of our agorithm. The fina poygona map in (c), obtained after 6 iterations, is composed of 86 segments, i.e., of 172 points. Thus, the proposed approach yieds the data compression ratio of 829:1. The mean distance of scan points to the cosest ine segments is 3.cm. We seected this map, since it contains surfaces of curved objects. The obtained poyines in (c) iustrate that the proposed approach is we suited to approximate inear as we as curved surfaces. For more detais on the appication of the proposed method in robot mapping see [14] (a) 1 1 (b) 1 1 (c) Fig. 4. (a) An origina outdoor map is composed of 142,9 scan points obtained during the Rescue Robot Camp in Rome, 24. We begin the approximation process with 192 ine segments that form the grid. (b) shows the output after the first iteration of our agorithm with 96 segments. (c) The fina poygona map obtained after 6 iterations is composed of ony 86 segments. The obtained compression rate is 829:1, and the approximation accuracy is 3.cm. Exampes iustrating fitting panar patches to 3D range data are given in [1]. Here we show ony one exampe in Fig.. Fig. (a) shows a 3D projection of the surface of some industria part composed of 744,4 aser range scan points, obtained from Fig. (b) shows
14 our approximation with 27 panar patches. The mean distance of each point to the cosest surface patch is.49 with the origina object size of (a) (b) Fig.. (a) An input surface with 744,4 sampe points. (b) Approximation with 27 panar patches. 7 Concusions The combination of Expectation Maximization Segment Fitting with aternating Segment Spitting and Merging was proven to be a powerfu too to gain a poyine representation of edge points in digita images, eading to a geometricay higher representation and an exceent data compression rate. The newy introduced, perceptua grouping based merging step baances the number of segments, created by partitioning and spitting, in a visuay natura way and therefore aows for the number of starting segments for the EM step to be imprecise. The extended EM agorithm is proven to yied gobay optima resuts. Acknowedgment This work was supported in part by the Nationa Science Foundation under grant NSF IIS We aso woud ike to acknowedge the support from the Nationa Institute of Standards and Technoogy, award Nr. NIST 7NANBH We thank Giorgio Grisetti (Univ. La Sapienza, Rome) for providing us the scan data. References 1. Bea, M.J., Ghahramani, Z.: The variationa bayesian em agorithm for incompete data. In: BAYESIAN STATISTICS 7. Oxford Univ. Press (23)
15 2. Wertheimer, M.: Untersuchungen zur ehre von der gestat ii. Psycoogische Forschung 4 (1923) Rosin, P.L.: Techniques for assessing poygona approximations of curves. IEEE Trans. PAMI 19(3) (1997) Ramer, U.: An iterative procedure for the poygona approximation of pane curves. Computer Graphics and Image Processing 1 (1972) Latecki, L.J., Lakämper, R.: Shape simiarity measure based on correspondence of visua parts. IEEE Trans. Pattern Anaysis and Machine Inteigence 22(1) (2) Latecki, L.J., Lakämper, R., Eckhardt, U.: Shape descriptors for non-rigid shapes with a singe cosed contour. In: Proc. of IEEE Conf. on Computer Vision and Pattern Recognition. (South Caroina, June 2) Sack, D., Burgard, W.: A comparison of methods for ine extraction from range data. In: Proc. of the th IFAC Symposium on Inteigent Autonomous Vehices (IAV). (24) 8. Veeck, M., Burgard, W.: Learning poyine maps from range scan data acquired with mobie robots. In: Proc. of the IEEE/RSJ Internationa Conference on Inteigent Robots and Systems (IROS). (24) 9. Scott, D.W.: Mutivariate Density Estimation: Theory, Practice, and Visuaization. Wiey and Sons (1992) 1. Siverman, B.W.: Density Estimation for Statistics and Data Anaysis. Chapman and Ha (1986) 11. Nea, R., Hinton, G.: A view of the em agorithm that justifies incrementa, sparse, and other variants. In Jordan, M.I., ed.: Learning in Graphica Modes, Kuwer (1998) 12. Arias-Castro, E., Donoho, D.L., Huo, X.: Adaptive mutiscae detection of fiamentary structures embedded in a background of uniform random points. Annas of Statistics (to appear) 13. Grisetti, G., Stachniss, C., Burgard, W.: Improving grid-based sam with raobackweized partice fiters by adaptive proposas and seective resamping. In: ICRA. (2) 14. Latecki, L.J., Lakaemper, R.: Poygona approximation of aser range data based on perceptua grouping and em. In: IEEE Int. Conf. on Robotics and Automation (ICRA). (26) 1. Lakaemper, R., Latecki, L.J.: Decomposition of 3d aser range data using panar patches. In: IEEE Int. Conf. on Robotics and Automation (ICRA). (26)
Language Identification for Texts Written in Transliteration
Language Identification for Texts Written in Transiteration Andrey Chepovskiy, Sergey Gusev, Margarita Kurbatova Higher Schoo of Economics, Data Anaysis and Artificia Inteigence Department, Pokrovskiy
More informationMobile App Recommendation: Maximize the Total App Downloads
Mobie App Recommendation: Maximize the Tota App Downoads Zhuohua Chen Schoo of Economics and Management Tsinghua University chenzhh3.12@sem.tsinghua.edu.cn Yinghui (Catherine) Yang Graduate Schoo of Management
More informationImage Segmentation Using Semi-Supervised k-means
I J C T A, 9(34) 2016, pp. 595-601 Internationa Science Press Image Segmentation Using Semi-Supervised k-means Reza Monsefi * and Saeed Zahedi * ABSTRACT Extracting the region of interest is a very chaenging
More informationA Design Method for Optimal Truss Structures with Certain Redundancy Based on Combinatorial Rigidity Theory
0 th Word Congress on Structura and Mutidiscipinary Optimization May 9 -, 03, Orando, Forida, USA A Design Method for Optima Truss Structures with Certain Redundancy Based on Combinatoria Rigidity Theory
More informationNearest Neighbor Learning
Nearest Neighbor Learning Cassify based on oca simiarity Ranges from simpe nearest neighbor to case-based and anaogica reasoning Use oca information near the current query instance to decide the cassification
More informationMulti-Robot Pose Graph Localization and Data Association from Unknown Initial Relative Poses
1 Muti-Robot Pose Graph Locaization and Data Association from Unknown Initia Reative Poses Vadim Indeman, Erik Neson, Nathan Michae and Frank Deaert Institute of Robotics and Inteigent Machines (IRIM)
More informationAN EVOLUTIONARY APPROACH TO OPTIMIZATION OF A LAYOUT CHART
13 AN EVOLUTIONARY APPROACH TO OPTIMIZATION OF A LAYOUT CHART Eva Vona University of Ostrava, 30th dubna st. 22, Ostrava, Czech Repubic e-mai: Eva.Vona@osu.cz Abstract: This artice presents the use of
More informationA Petrel Plugin for Surface Modeling
A Petre Pugin for Surface Modeing R. M. Hassanpour, S. H. Derakhshan and C. V. Deutsch Structure and thickness uncertainty are important components of any uncertainty study. The exact ocations of the geoogica
More informationA Two-Step Approach to Hallucinating Faces: Global Parametric Model and Local Nonparametric Model
A Two-Step Approach to aucinating Faces: Goba Parametric Mode and Loca Nonparametric Mode Ce Liu eung-yeung Shum Chang-Shui Zhang State Key Lab of nteigent Technoogy and Systems, Dept. of Automation, Tsinghua
More informationACTIVE LEARNING ON WEIGHTED GRAPHS USING ADAPTIVE AND NON-ADAPTIVE APPROACHES. Eyal En Gad, Akshay Gadde, A. Salman Avestimehr and Antonio Ortega
ACTIVE LEARNING ON WEIGHTED GRAPHS USING ADAPTIVE AND NON-ADAPTIVE APPROACHES Eya En Gad, Akshay Gadde, A. Saman Avestimehr and Antonio Ortega Department of Eectrica Engineering University of Southern
More informationNeural Network Enhancement of the Los Alamos Force Deployment Estimator
Missouri University of Science and Technoogy Schoars' Mine Eectrica and Computer Engineering Facuty Research & Creative Works Eectrica and Computer Engineering 1-1-1994 Neura Network Enhancement of the
More informationSensitivity Analysis of Hopfield Neural Network in Classifying Natural RGB Color Space
Sensitivity Anaysis of Hopfied Neura Network in Cassifying Natura RGB Coor Space Department of Computer Science University of Sharjah UAE rsammouda@sharjah.ac.ae Abstract: - This paper presents a study
More informationA Memory Grouping Method for Sharing Memory BIST Logic
A Memory Grouping Method for Sharing Memory BIST Logic Masahide Miyazai, Tomoazu Yoneda, and Hideo Fuiwara Graduate Schoo of Information Science, Nara Institute of Science and Technoogy (NAIST), 8916-5
More informationBacking-up Fuzzy Control of a Truck-trailer Equipped with a Kingpin Sliding Mechanism
Backing-up Fuzzy Contro of a Truck-traier Equipped with a Kingpin Siding Mechanism G. Siamantas and S. Manesis Eectrica & Computer Engineering Dept., University of Patras, Patras, Greece gsiama@upatras.gr;stam.manesis@ece.upatras.gr
More informationA Fast Block Matching Algorithm Based on the Winner-Update Strategy
In Proceedings of the Fourth Asian Conference on Computer Vision, Taipei, Taiwan, Jan. 000, Voume, pages 977 98 A Fast Bock Matching Agorithm Based on the Winner-Update Strategy Yong-Sheng Chenyz Yi-Ping
More informationFiltering. Yao Wang Polytechnic University, Brooklyn, NY 11201
Spatia Domain Linear Fitering Yao Wang Poytechnic University Brookyn NY With contribution rom Zhu Liu Onur Gueryuz and Gonzaez/Woods Digita Image Processing ed Introduction Outine Noise remova using ow-pass
More informationMulti-level Shape Recognition based on Wavelet-Transform. Modulus Maxima
uti-eve Shape Recognition based on Waveet-Transform oduus axima Faouzi Aaya Cheikh, Azhar Quddus and oncef Gabbouj Tampere University of Technoogy (TUT), Signa Processing aboratory, P.O. Box 553, FIN-33101
More informationA METHOD FOR GRIDLESS ROUTING OF PRINTED CIRCUIT BOARDS. A. C. Finch, K. J. Mackenzie, G. J. Balsdon, G. Symonds
A METHOD FOR GRIDLESS ROUTING OF PRINTED CIRCUIT BOARDS A C Finch K J Mackenzie G J Basdon G Symonds Raca-Redac Ltd Newtown Tewkesbury Gos Engand ABSTRACT The introduction of fine-ine technoogies to printed
More informationCommunity-Aware Opportunistic Routing in Mobile Social Networks
IEEE TRANSACTIONS ON COMPUTERS VOL:PP NO:99 YEAR 213 Community-Aware Opportunistic Routing in Mobie Socia Networks Mingjun Xiao, Member, IEEE Jie Wu, Feow, IEEE, and Liusheng Huang, Member, IEEE Abstract
More informationJoint disparity and motion eld estimation in. stereoscopic image sequences. Ioannis Patras, Nikos Alvertos and Georgios Tziritas y.
FORTH-ICS / TR-157 December 1995 Joint disparity and motion ed estimation in stereoscopic image sequences Ioannis Patras, Nikos Avertos and Georgios Tziritas y Abstract This work aims at determining four
More informationUniversity of Illinois at Urbana-Champaign, Urbana, IL 61801, /11/$ IEEE 162
oward Efficient Spatia Variation Decomposition via Sparse Regression Wangyang Zhang, Karthik Baakrishnan, Xin Li, Duane Boning and Rob Rutenbar 3 Carnegie Meon University, Pittsburgh, PA 53, wangyan@ece.cmu.edu,
More informationM. Badent 1, E. Di Giacomo 2, G. Liotta 2
DIEI Dipartimento di Ingegneria Eettronica e de informazione RT 005-06 Drawing Coored Graphs on Coored Points M. Badent 1, E. Di Giacomo 2, G. Liotta 2 1 University of Konstanz 2 Università di Perugia
More informationAn Adaptive Two-Copy Delayed SR-ARQ for Satellite Channels with Shadowing
An Adaptive Two-Copy Deayed SR-ARQ for Sateite Channes with Shadowing Jing Zhu, Sumit Roy zhuj@ee.washington.edu Department of Eectrica Engineering, University of Washington Abstract- The paper focuses
More informationA New Supervised Clustering Algorithm Based on Min-Max Modular Network with Gaussian-Zero-Crossing Functions
2006 Internationa Joint Conference on Neura Networks Sheraton Vancouver Wa Centre Hote, Vancouver, BC, Canada Juy 16-21, 2006 A New Supervised Custering Agorithm Based on Min-Max Moduar Network with Gaussian-Zero-Crossing
More informationChapter Multidimensional Direct Search Method
Chapter 09.03 Mutidimensiona Direct Search Method After reading this chapter, you shoud be abe to:. Understand the fundamentas of the mutidimensiona direct search methods. Understand how the coordinate
More informationDistance Weighted Discrimination and Second Order Cone Programming
Distance Weighted Discrimination and Second Order Cone Programming Hanwen Huang, Xiaosun Lu, Yufeng Liu, J. S. Marron, Perry Haaand Apri 3, 2012 1 Introduction This vignette demonstrates the utiity and
More informationAutomatic Grouping for Social Networks CS229 Project Report
Automatic Grouping for Socia Networks CS229 Project Report Xiaoying Tian Ya Le Yangru Fang Abstract Socia networking sites aow users to manuay categorize their friends, but it is aborious to construct
More informationBinarized support vector machines
Universidad Caros III de Madrid Repositorio instituciona e-archivo Departamento de Estadística http://e-archivo.uc3m.es DES - Working Papers. Statistics and Econometrics. WS 2007-11 Binarized support vector
More informationFurther Optimization of the Decoding Method for Shortened Binary Cyclic Fire Code
Further Optimization of the Decoding Method for Shortened Binary Cycic Fire Code Ch. Nanda Kishore Heosoft (India) Private Limited 8-2-703, Road No-12 Banjara His, Hyderabad, INDIA Phone: +91-040-3378222
More informationIdentifying and Tracking Pedestrians Based on Sensor Fusion and Motion Stability Predictions
Sensors 2010, 10, 8028-8053; doi:10.3390/s100908028 OPEN ACCESS sensors ISSN 1424-8220 www.mdpi.com/journa/sensors Artice Identifying and Tracking Pedestrians Based on Sensor Fusion and Motion Stabiity
More informationGeometric clustering for line drawing simplification
Eurographics Symposium on Rendering (2005) Kavita Baa, Phiip Dutré (Editors) Geometric custering for ine drawing simpification P. Bara, J.Thoot and F. X. Siion ε ARTIS GRAVIR/IMAG INRIA Figure 1: The two
More informationEndoscopic Motion Compensation of High Speed Videoendoscopy
Endoscopic Motion Compensation of High Speed Videoendoscopy Bharath avuri Department of Computer Science and Engineering, University of South Caroina, Coumbia, SC - 901. ravuri@cse.sc.edu Abstract. High
More informationHandling Outliers in Non-Blind Image Deconvolution
Handing Outiers in Non-Bind Image Deconvoution Sunghyun Cho 1 Jue Wang 2 Seungyong Lee 1,2 sodomau@postech.ac.kr juewang@adobe.com eesy@postech.ac.kr 1 POSTECH 2 Adobe Systems Abstract Non-bind deconvoution
More informationInnerSpec: Technical Report
InnerSpec: Technica Report Fabrizio Guerrini, Aessandro Gnutti, Riccardo Leonardi Department of Information Engineering, University of Brescia Via Branze 38, 25123 Brescia, Itay {fabrizio.guerrini, a.gnutti006,
More informationMultiple Plane Phase Retrieval Based On Inverse Regularized Imaging and Discrete Diffraction Transform
Mutipe Pane Phase Retrieva Based On Inverse Reguaried Imaging and Discrete Diffraction Transform Artem Migukin, Vadimir Katkovnik, and Jaakko Astoa Department of Signa Processing, Tampere University of
More informationOptimization and Application of Support Vector Machine Based on SVM Algorithm Parameters
Optimization and Appication of Support Vector Machine Based on SVM Agorithm Parameters YAN Hui-feng 1, WANG Wei-feng 1, LIU Jie 2 1 ChongQing University of Posts and Teecom 400065, China 2 Schoo Of Civi
More informationCLOUD RADIO ACCESS NETWORK WITH OPTIMIZED BASE-STATION CACHING
CLOUD RADIO ACCESS NETWORK WITH OPTIMIZED BASE-STATION CACHING Binbin Dai and Wei Yu Ya-Feng Liu Department of Eectrica and Computer Engineering University of Toronto, Toronto ON, Canada M5S 3G4 Emais:
More informationSubstitute Model of Deep-groove Ball Bearings in Numeric Analysis of Complex Constructions Like Manipulators
Mechanics and Mechanica Engineering Vo. 12, No. 4 (2008) 349 356 c Technica University of Lodz Substitute Mode of Deep-groove Ba Bearings in Numeric Anaysis of Compex Constructions Like Manipuators Leszek
More informationOutline. Parallel Numerical Algorithms. Forward Substitution. Triangular Matrices. Solving Triangular Systems. Back Substitution. Parallel Algorithm
Outine Parae Numerica Agorithms Chapter 8 Prof. Michae T. Heath Department of Computer Science University of Iinois at Urbana-Champaign CS 554 / CSE 512 1 2 3 4 Trianguar Matrices Michae T. Heath Parae
More informationQuality Assessment using Tone Mapping Algorithm
Quaity Assessment using Tone Mapping Agorithm Nandiki.pushpa atha, Kuriti.Rajendra Prasad Research Schoar, Assistant Professor, Vignan s institute of engineering for women, Visakhapatnam, Andhra Pradesh,
More informationComputer Graphics. - Shading & Texturing -
Computer Graphics - Shading & Texturing - Empirica BRDF Approximation Purey heuristic mode Initiay without units (vaues [0,1] r = r,a + r,d + r,s ( + r,m + r,t r,a : Ambient term Approximate indirect iumination
More informationMulti-task hidden Markov modeling of spectrogram feature from radar high-resolution range profiles
http://asp.eurasipjournas.com/content/22//86 RESEARCH Open Access Muti-task hidden Markov modeing of spectrogram feature from radar high-resoution range profies Mian Pan, Lan Du *, Penghui Wang, Hongwei
More informationEfficient Histogram-based Indexing for Video Copy Detection
Efficient Histogram-based Indexing for Video Copy Detection Chih-Yi Chiu, Jenq-Haur Wang*, and Hung-Chi Chang Institute of Information Science, Academia Sinica, Taiwan *Department of Computer Science and
More informationResource Optimization to Provision a Virtual Private Network Using the Hose Model
Resource Optimization to Provision a Virtua Private Network Using the Hose Mode Monia Ghobadi, Sudhakar Ganti, Ghoamai C. Shoja University of Victoria, Victoria C, Canada V8W 3P6 e-mai: {monia, sganti,
More informationSpace-Time Trade-offs.
Space-Time Trade-offs. Chethan Kamath 03.07.2017 1 Motivation An important question in the study of computation is how to best use the registers in a CPU. In most cases, the amount of registers avaiabe
More informationAn Exponential Time 2-Approximation Algorithm for Bandwidth
An Exponentia Time 2-Approximation Agorithm for Bandwidth Martin Fürer 1, Serge Gaspers 2, Shiva Prasad Kasiviswanathan 3 1 Computer Science and Engineering, Pennsyvania State University, furer@cse.psu.edu
More informationTopology-aware Key Management Schemes for Wireless Multicast
Topoogy-aware Key Management Schemes for Wireess Muticast Yan Sun, Wade Trappe,andK.J.RayLiu Department of Eectrica and Computer Engineering, University of Maryand, Coege Park Emai: ysun, kjriu@gue.umd.edu
More informationSequential Learning of Layered Models from Video
Sequentia Learning of Layered Modes from Video Michais K. Titsias and Christopher K. I. Wiiams Schoo of Informatics, University of Edinburgh, Edinburgh EH1 2QL, UK M.Titsias@sms.ed.ac.uk, c.k.i.wiiams@ed.ac.uk
More information(0,l) (0,0) (l,0) (l,l)
Parae Domain Decomposition and Load Baancing Using Space-Fiing Curves Srinivas Auru Fatih E. Sevigen Dept. of CS Schoo of EECS New Mexico State University Syracuse University Las Cruces, NM 88003-8001
More informationJOINT IMAGE REGISTRATION AND EXAMPLE-BASED SUPER-RESOLUTION ALGORITHM
JOINT IMAGE REGISTRATION AND AMPLE-BASED SUPER-RESOLUTION ALGORITHM Hyo-Song Kim, Jeyong Shin, and Rae-Hong Park Department of Eectronic Engineering, Schoo of Engineering, Sogang University 35 Baekbeom-ro,
More informationApplication of Intelligence Based Genetic Algorithm for Job Sequencing Problem on Parallel Mixed-Model Assembly Line
American J. of Engineering and Appied Sciences 3 (): 5-24, 200 ISSN 94-7020 200 Science Pubications Appication of Inteigence Based Genetic Agorithm for Job Sequencing Probem on Parae Mixed-Mode Assemby
More informationTransformation Invariance in Pattern Recognition: Tangent Distance and Propagation
Transformation Invariance in Pattern Recognition: Tangent Distance and Propagation Patrice Y. Simard, 1 Yann A. Le Cun, 2 John S. Denker, 2 Bernard Victorri 3 1 Microsoft Research, 1 Microsoft Way, Redmond,
More informationUtility-based Camera Assignment in a Video Network: A Game Theoretic Framework
This artice has been accepted for pubication in a future issue of this journa, but has not been fuy edited. Content may change prior to fina pubication. Y.LI AND B.BHANU CAMERA ASSIGNMENT: A GAME-THEORETIC
More informationLecture Notes for Chapter 4 Part III. Introduction to Data Mining
Data Mining Cassification: Basic Concepts, Decision Trees, and Mode Evauation Lecture Notes for Chapter 4 Part III Introduction to Data Mining by Tan, Steinbach, Kumar Adapted by Qiang Yang (2010) Tan,Steinbach,
More informationAutomatic Hidden Web Database Classification
Automatic idden Web atabase Cassification Zhiguo Gong, Jingbai Zhang, and Qian Liu Facuty of Science and Technoogy niversity of Macau Macao, PRC {fstzgg,ma46597,ma46620}@umac.mo Abstract. In this paper,
More informationAs Michi Henning and Steve Vinoski showed 1, calling a remote
Reducing CORBA Ca Latency by Caching and Prefetching Bernd Brügge and Christoph Vismeier Technische Universität München Method ca atency is a major probem in approaches based on object-oriented middeware
More informationOn Upper Bounds for Assortment Optimization under the Mixture of Multinomial Logit Models
On Upper Bounds for Assortment Optimization under the Mixture of Mutinomia Logit Modes Sumit Kunnumka September 30, 2014 Abstract The assortment optimization probem under the mixture of mutinomia ogit
More informationOptimized Base-Station Cache Allocation for Cloud Radio Access Network with Multicast Backhaul
Optimized Base-Station Cache Aocation for Coud Radio Access Network with Muticast Backhau Binbin Dai, Student Member, IEEE, Ya-Feng Liu, Member, IEEE, and Wei Yu, Feow, IEEE arxiv:804.0730v [cs.it] 28
More informationResponse Surface Model Updating for Nonlinear Structures
Response Surface Mode Updating for Noninear Structures Gonaz Shahidi a, Shamim Pakzad b a PhD Student, Department of Civi and Environmenta Engineering, Lehigh University, ATLSS Engineering Research Center,
More informationSolving Large Double Digestion Problems for DNA Restriction Mapping by Using Branch-and-Bound Integer Linear Programming
The First Internationa Symposium on Optimization and Systems Bioogy (OSB 07) Beijing, China, August 8 10, 2007 Copyright 2007 ORSC & APORC pp. 267 279 Soving Large Doube Digestion Probems for DNA Restriction
More informationSemi-Supervised Learning with Sparse Distributed Representations
Semi-Supervised Learning with Sparse Distributed Representations David Zieger dzieger@stanford.edu CS 229 Fina Project 1 Introduction For many machine earning appications, abeed data may be very difficut
More informationSequential Learning of Layered Models from Video
Sequentia Learning of Layered Modes from Video Michais K. Titsias and Christopher K. I. Wiiams Schoo of Informatics, University of Edinburgh, Edinburgh EH1 2QL, UK M.Titsias@sms.ed.ac.uk, c.k.i.wiiams@ed.ac.uk
More informationA HYBRID FEATURE SELECTION METHOD BASED ON FISHER SCORE AND GENETIC ALGORITHM
Journa of Mathematica Sciences: Advances and Appications Voume 37, 2016, Pages 51-78 Avaiabe at http://scientificadvances.co.in DOI: http://dx.doi.org/10.18642/jmsaa_7100121627 A HYBRID FEATURE SELECTION
More informationFastest-Path Computation
Fastest-Path Computation DONGHUI ZHANG Coege of Computer & Information Science Northeastern University Synonyms fastest route; driving direction Definition In the United states, ony 9.% of the househods
More informationArithmetic Coding. Prof. Ja-Ling Wu. Department of Computer Science and Information Engineering National Taiwan University
Arithmetic Coding Prof. Ja-Ling Wu Department of Computer Science and Information Engineering Nationa Taiwan University F(X) Shannon-Fano-Eias Coding W..o.g. we can take X={,,,m}. Assume p()>0 for a. The
More informationPrivacy Preserving Subgraph Matching on Large Graphs in Cloud
Privacy Preserving Subgraph Matching on Large Graphs in Coud Zhao Chang,#, Lei Zou, Feifei Li # Peing University, China; # University of Utah, USA; {changzhao,zouei}@pu.edu.cn; {zchang,ifeifei}@cs.utah.edu
More informationAN OVERVIEW OF IMAGE SEGMENTATION TECHNIQUES. IN Fabris-Rotelli 1 and J-F Greeff 1. ABSTRACT
AN OVERVIEW OF IMAGE SEGMENTATION TECHNIQUES IN Fabris-Rotei 1 and J-F Greeff 1 1 Department of Statistics, University of Pretoria, 0002, Pretoria, South Africa, inger.fabris-rotei@up.ac.za ABSTRACT We
More informationSolutions to the Final Exam
CS/Math 24: Intro to Discrete Math 5//2 Instructor: Dieter van Mekebeek Soutions to the Fina Exam Probem Let D be the set of a peope. From the definition of R we see that (x, y) R if and ony if x is a
More informationAlpha labelings of straight simple polyominal caterpillars
Apha abeings of straight simpe poyomina caterpiars Daibor Froncek, O Nei Kingston, Kye Vezina Department of Mathematics and Statistics University of Minnesota Duuth University Drive Duuth, MN 82-3, U.S.A.
More informationLecture outline Graphics and Interaction Scan Converting Polygons and Lines. Inside or outside a polygon? Scan conversion.
Lecture outine 433-324 Graphics and Interaction Scan Converting Poygons and Lines Department of Computer Science and Software Engineering The Introduction Scan conversion Scan-ine agorithm Edge coherence
More informationGeneralizing the Dubins and Reeds-Shepp cars: fastest paths for bounded-velocity mobile robots
2008 IEEE Internationa Conference on Robotics and Automation Pasadena, CA, USA, May 19-23, 2008 Generaizing the Dubins and Reeds-Shepp cars: fastest paths for bounded-veocity mobie robots Andrei A. Furtuna,
More informationSpecial Edition Using Microsoft Excel Selecting and Naming Cells and Ranges
Specia Edition Using Microsoft Exce 2000 - Lesson 3 - Seecting and Naming Ces and.. Page 1 of 8 [Figures are not incuded in this sampe chapter] Specia Edition Using Microsoft Exce 2000-3 - Seecting and
More informationDETECTION OF OBSTACLE AND FREESPACE IN AN AUTONOMOUS WHEELCHAIR USING A STEREOSCOPIC CAMERA SYSTEM
DETECTION OF OBSTACLE AND FREESPACE IN AN AUTONOMOUS WHEELCHAIR USING A STEREOSCOPIC CAMERA SYSTEM Le Minh 1, Thanh Hai Nguyen 2, Tran Nghia Khanh 2, Vo Văn Toi 2, Ngo Van Thuyen 1 1 University of Technica
More informationTechTest2017. Solutions Key. Final Edit Copy. Merit Scholarship Examination in the Sciences and Mathematics given on 1 April 2017, and.
TechTest07 Merit Schoarship Examination in the Sciences and Mathematics given on Apri 07, and sponsored by The Sierra Economics and Science Foundation Soutions Key V9feb7 TechTest07 Soutions Key / 9 07
More informationMACHINE learning techniques can, automatically,
Proceedings of Internationa Joint Conference on Neura Networks, Daas, Texas, USA, August 4-9, 203 High Leve Data Cassification Based on Network Entropy Fiipe Aves Neto and Liang Zhao Abstract Traditiona
More informationCERIAS Tech Report Replicated Parallel I/O without Additional Scheduling Costs by Mikhail J. Atallah Center for Education and Research
CERIAS Tech Report 2003-50 Repicated Parae I/O without Additiona Scheduing Costs by Mikhai J. Ataah Center for Education and Research Information Assurance and Security Purdue University, West Lafayette,
More informationDigital Image Watermarking Algorithm Based on Fast Curvelet Transform
J. Software Engineering & Appications, 010, 3, 939-943 doi:10.436/jsea.010.310111 Pubished Onine October 010 (http://www.scirp.org/journa/jsea) 939 igita Image Watermarking Agorithm Based on Fast Curveet
More informationTHE PERCENTAGE OCCUPANCY HIT OR MISS TRANSFORM
17th European Signa Processing Conference (EUSIPCO 2009) Gasgow, Scotand, August 24-28, 2009 THE PERCENTAGE OCCUPANCY HIT OR MISS TRANSFORM P. Murray 1, S. Marsha 1, and E.Buinger 2 1 Dept. of Eectronic
More informationStereo. CS 510 May 2 nd, 2014
Stereo CS 510 May 2 nd, 2014 Where are we? We are done! (essentiay) We covered image matching Correation & Correation Fiters Fourier Anaysis PCA We covered feature-based matching Bag of Features approach
More informationA Comparison of a Second-Order versus a Fourth- Order Laplacian Operator in the Multigrid Algorithm
A Comparison of a Second-Order versus a Fourth- Order Lapacian Operator in the Mutigrid Agorithm Kaushik Datta (kdatta@cs.berkeey.edu Math Project May 9, 003 Abstract In this paper, the mutigrid agorithm
More informationUnsupervised Segmentation of Objects using Efficient Learning
Unsupervised Segmentation of Objects using Efficient Learning {Himanshu Arora, Nicoas Loeff}, David A. Forsyth, Narendra Ahuja University of Iinois at Urbana-Champaign, Urbana, IL, 60 {harora,oeff,daf,n-ahuja}@uiuc.edu
More informationfastruct: model-based clustering made faster
fastruct: mode-based custering made faster Chibiao Chen Forence Forbes Oivier François INRIA Rhone-Apes, 655 Avenue de Europe, Montbonnot, 38334 St Ismier France TIMC-TIMB, Dept Math Bioogy, Facuty of
More informationA Robust Sign Language Recognition System with Sparsely Labeled Instances Using Wi-Fi Signals
A Robust Sign Language Recognition System with Sparsey Labeed Instances Using Wi-Fi Signas Jiacheng Shang, Jie Wu Center for Networked Computing Dept. of Computer and Info. Sciences Tempe University Motivation
More informationCHAPTER 5 EXPERIMENTAL RESULTS. 5.1 Boresight Calibration
CHAPTER 5 EXPERIMENTAL RESULTS 5. Boresight Caibration To compare the accuracy of determined boresight misaignment parameters, both the manua tie point seections and the tie points detected with image
More informationDevelopment of a hybrid K-means-expectation maximization clustering algorithm
Journa of Computations & Modeing, vo., no.4, 0, -3 ISSN: 79-765 (print, 79-8850 (onine Scienpress Ltd, 0 Deveopment of a hybrid K-means-expectation maximization custering agorithm Adigun Abimboa Adebisi,
More informationAUTOMATIC IMAGE RETARGETING USING SALIENCY BASED MESH PARAMETERIZATION
S.Sai Kumar et a. / (IJCSIT Internationa Journa of Computer Science and Information Technoogies, Vo. 1 (4, 010, 73-79 AUTOMATIC IMAGE RETARGETING USING SALIENCY BASED MESH PARAMETERIZATION 1 S.Sai Kumar,
More informationOn-Chip CNN Accelerator for Image Super-Resolution
On-Chip CNN Acceerator for Image Super-Resoution Jung-Woo Chang and Suk-Ju Kang Dept. of Eectronic Engineering, Sogang University, Seou, South Korea {zwzang91, sjkang}@sogang.ac.kr ABSTRACT To impement
More informationHiding secrete data in compressed images using histogram analysis
University of Woongong Research Onine University of Woongong in Dubai - Papers University of Woongong in Dubai 2 iding secrete data in compressed images using histogram anaysis Farhad Keissarian University
More informationResearch of Classification based on Deep Neural Network
2018 Internationa Conference on Sensor Network and Computer Engineering (ICSNCE 2018) Research of Emai Cassification based on Deep Neura Network Wang Yawen Schoo of Computer Science and Engineering Xi
More informationA Column Generation Approach for Support Vector Machines
A Coumn Generation Approach for Support Vector Machines Emiio Carrizosa Universidad de Sevia (Spain). ecarrizosa@us.es Beén Martín-Barragán Universidad de Sevia (Spain). bemart@us.es Doores Romero Moraes
More informationWATERMARKING GIS DATA FOR DIGITAL MAP COPYRIGHT PROTECTION
WATERMARKING GIS DATA FOR DIGITAL MAP COPYRIGHT PROTECTION Shen Tao Chinese Academy of Surveying and Mapping, Beijing 100039, China shentao@casm.ac.cn Xu Dehe Institute of resources and environment, North
More informationEighth Annual ACM-SIAM Symposium on Discrete Algorithms, January 1997, pp. 777{ Partial Matching of Planar Polylines
Eighth Annua ACM-SIAM Symposium on Discrete Agorithm January 997, pp. 777{786 Partia Matching of Panar Poyines Under Simiarity Transformations Scott D. Cohen Leonidas J. Guibas fscohen,guibasg@cs.stanford.edu
More informationHierarchical Volumetric Multi-view Stereo Reconstruction of Manifold Surfaces based on Dual Graph Embedding
Hierarchica Voumetric Muti-view Stereo Reconstruction of Manifod Surfaces based on Dua Graph Embedding Aexander Hornung and Leif Kobbet Computer Graphics Group, RWTH Aachen University http://www.rwth-graphics.de
More informationPriority Queueing for Packets with Two Characteristics
1 Priority Queueing for Packets with Two Characteristics Pave Chuprikov, Sergey I. Nikoenko, Aex Davydow, Kiri Kogan Abstract Modern network eements are increasingy required to dea with heterogeneous traffic.
More informationFurther Concepts in Geometry
ppendix F Further oncepts in Geometry F. Exporing ongruence and Simiarity Identifying ongruent Figures Identifying Simiar Figures Reading and Using Definitions ongruent Trianges assifying Trianges Identifying
More informationMULTITASK MULTIVARIATE COMMON SPARSE REPRESENTATIONS FOR ROBUST MULTIMODAL BIOMETRICS RECOGNITION. Heng Zhang, Vishal M. Patel and Rama Chellappa
MULTITASK MULTIVARIATE COMMON SPARSE REPRESENTATIONS FOR ROBUST MULTIMODAL BIOMETRICS RECOGNITION Heng Zhang, Visha M. Pate and Rama Cheappa Center for Automation Research University of Maryand, Coage
More informationFREE-FORM ANISOTROPY: A NEW METHOD FOR CRACK DETECTION ON PAVEMENT SURFACE IMAGES
FREE-FORM ANISOTROPY: A NEW METHOD FOR CRACK DETECTION ON PAVEMENT SURFACE IMAGES Tien Sy Nguyen, Stéphane Begot, Forent Ducuty, Manue Avia To cite this version: Tien Sy Nguyen, Stéphane Begot, Forent
More informationCrossing Minimization Problems of Drawing Bipartite Graphs in Two Clusters
Crossing Minimiation Probems o Drawing Bipartite Graphs in Two Custers Lanbo Zheng, Le Song, and Peter Eades Nationa ICT Austraia, and Schoo o Inormation Technoogies, University o Sydney,Austraia Emai:
More informationFuzzy Equivalence Relation Based Clustering and Its Use to Restructuring Websites Hyperlinks and Web Pages
Fuzzy Equivaence Reation Based Custering and Its Use to Restructuring Websites Hyperinks and Web Pages Dimitris K. Kardaras,*, Xenia J. Mamakou, and Bi Karakostas 2 Business Informatics Laboratory, Dept.
More informationAn Alternative Approach for Solving Bi-Level Programming Problems
American Journa of Operations Research, 07, 7, 9-7 http://www.scirp.org/ourna/aor ISSN Onine: 60-889 ISSN rint: 60-880 An Aternative Approach for Soving Bi-Leve rogramming robems Rashmi Bira, Viay K. Agarwa,
More information