iccv iccv2013 iccv2013-283 knowledge-graph by maker-knowledge-mining

283 iccv-2013-Multiple Non-rigid Surface Detection and Registration


Source: pdf

Author: Yi Wu, Yoshihisa Ijiri, Ming-Hsuan Yang

Abstract: Detecting and registering nonrigid surfaces are two important research problems for computer vision. Much work has been done with the assumption that there exists only one instance in the image. In this work, we propose an algorithm that detects and registers multiple nonrigid instances of given objects in a cluttered image. Specifically, after we use low level feature points to obtain the initial matches between templates and the input image, a novel high-order affinity graph is constructed to model the consistency of local topology. A hierarchical clustering approach is then used to locate the nonrigid surfaces. To remove the outliers in the cluster, we propose a deterministic annealing approach based on the Thin Plate Spline (TPS) model. The proposed method achieves high accuracy even when the number of outliers is nineteen times larger than the inliers. As the matches may appear sparsely in each instance, we propose a TPS based match growing approach to propagate the matches. Finally, an approach that fuses feature and appearance information is proposed to register each nonrigid surface. Extensive experiments and evaluations demonstrate that the proposed algorithm achieves promis- ing results in detecting and registering multiple non-rigid surfaces in a cluttered scene.

Reference: text


Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 edu Abstract Detecting and registering nonrigid surfaces are two important research problems for computer vision. [sent-7, score-0.255]

2 Specifically, after we use low level feature points to obtain the initial matches between templates and the input image, a novel high-order affinity graph is constructed to model the consistency of local topology. [sent-10, score-0.445]

3 A hierarchical clustering approach is then used to locate the nonrigid surfaces. [sent-11, score-0.256]

4 To remove the outliers in the cluster, we propose a deterministic annealing approach based on the Thin Plate Spline (TPS) model. [sent-12, score-0.532]

5 As the matches may appear sparsely in each instance, we propose a TPS based match growing approach to propagate the matches. [sent-14, score-0.409]

6 Finally, an approach that fuses feature and appearance information is proposed to register each nonrigid surface. [sent-15, score-0.328]

7 This is a difficult problem since the appearance of imaged surfaces varies due to many factors such as camera pose, surface deformation and lighting conditions. [sent-19, score-0.36]

8 When multiple deformed instances of an object appear in an image, this makes the detection and registration task even more challenging. [sent-21, score-0.348]

9 In this paper, we address the multiple non-rigid surface detection and registration problem (See Figure 1). [sent-29, score-0.317]

10 Given the template image representing the surface of interest in a canonical shape, the goal is to locate the deformed instances of the surface and estimate the deformation parameters of each instance. [sent-30, score-0.681]

11 , SIFT [18]) to obtain the feature points in both template and input image, and then the correspondences between these two sets are constructed based on the similarity of local appearance around feature points. [sent-33, score-0.501]

12 After rejecting outliers in each correspondence cluster, the sparsely distributed matches are propagated by a Thin Plate Spline (TPS) based match growing algorithm. [sent-35, score-0.68]

13 The deformation parameters of each nonrigid surface are estimated by an approach that fuses feature and appearance information based on the TPS model. [sent-36, score-0.506]

14 (a) Two templates; (b) The detection and registration results overlaid on the input image, where the estimated non-rigid warps are illustrated in deformable meshes with different colors for different objects. [sent-40, score-0.34]

15 The feature-based approaches rely on establishing correspondences between features of a template and an input image. [sent-44, score-0.286]

16 On the other hand, the appearance-based approaches do not rely on features and directly minimize the intensity discrepancy between the template and an input image warped back onto the coordinate frame of the template, which could achieve better registration accuracy. [sent-46, score-0.421]

17 The feature-based approaches first establish correspondences by feature matching, then after eliminating outliers the transformation is estimated. [sent-49, score-0.413]

18 With 25% of outliers and 100 transformation parameters, which is a common case for nonrigid surface registration, a RANSAC approach requires 1012 samples to guarantee that, with high probability, at least one sample is included without outliers [15]. [sent-55, score-0.681]

19 [23] adopt a robust estimator to measure the outliers and employ an iterative semi-implicit optimization scheme to refine the matches. [sent-57, score-0.279]

20 In [10], a robust point set matching approach, TPS-RPM, is proposed to match points and remove outlier points iteratively. [sent-60, score-0.456]

21 The TPS-RPM method solves the joint optimization problem iteratively by deterministic annealing and soft assignment with one-toone correspondence assumption, which is computationally expensive. [sent-61, score-0.397]

22 In this paper, we adopt the deterministic annealing approach, but use it only to remove spurious matches. [sent-62, score-0.364]

23 Recent feature matching methods formulate visual correspondence as a graph matching problem by considering pairwise geometric distortion of objects between im- ages [16, 4, 25]. [sent-65, score-0.278]

24 A novel high-order affinity graph is constructed to model the consistency of local topology, and a hierarchical clustering approach is used to locate the nonrigid surfaces. [sent-79, score-0.38]

25 Both approaches depend on the affine transformation from the affine feature detector [20] for match growing. [sent-84, score-0.355]

26 In this work, to deal with the case that the matches are too sparsely distributed, we propose an approach to propagate matches locally by using the TPS model. [sent-85, score-0.435]

27 Matthews and Baker [21] improves the active appearance model with an inverse composition algorithm [1] in which triangulation meshes are used and piecewise affine warps are estimated. [sent-91, score-0.269]

28 The piecewise affine warp is used to map points, where the mapping of one point only considers three surrounding control points. [sent-95, score-0.335]

29 In this paper, we propose a fusion approach using the TPS warp which maps the points in a holistic way. [sent-96, score-0.349]

30 The fusion approach can deal with large distortions in which correct feature correspondences are difficult to obtain. [sent-97, score-0.274]

31 The initial feature correspondences between a template and an input image are obtained by comparing the SIFT descriptors based on Euclidean distance using the k nearest-neighbor algorithm. [sent-100, score-0.325]

32 Note that each feature point in the template may be matched to multiple points in the input image. [sent-101, score-0.288]

33 As it requires at least three correspondences to determine one affine transformation, we construct a high order graph, G = (V, E), of triple correspondences. [sent-112, score-0.246]

34 We can also consider G as a graph of affine transformations and each node is an affine matrix. [sent-114, score-0.261]

35 However, most of these CTPs contain false correspondences and we propose a filter based on local topology to reject most outlier CTPs before constructing the high order graph. [sent-117, score-0.393]

36 We compute the attribute by using the affine matrix to transform each triangle in the template to the input image plane and compute the Euclidean distances between the transformed points and the corresponding points in the input image. [sent-130, score-0.425]

37 A few trivial clusters th,at C co=nta {inM small number) o}f matches are classified as outliers, and others correspond to instances of the template object. [sent-150, score-0.388]

38 As the correspondences in a cluster may still contain some outliers, we propose a fast determinative annealing method based on the TPS deformation model to further refine the matches. [sent-153, score-0.525]

39 In the correspondence cluster i,the sum of squared residuals of Ni correspondences based on the TPS model is: Ef=N1in? [sent-255, score-0.271]

40 We propose an approach based on the deterministic annealing algorithm [27] to gradually reduce λs. [sent-285, score-0.35]

41 At the beginning of the annealing process, λs is large, which greatly limits the range of deformation and avoids overfitting due to the outliers. [sent-286, score-0.387]

42 Furthermore, during the annealing process, the threshold, d, to determine the outliers is also gradually decreased. [sent-288, score-0.468]

43 er I wf dhere d is also gradually reduced via the annealing procedure by setting d = T · d0. [sent-303, score-0.264]

44 The non-rigid warps estimated by feature-based approach and fusion-based approach based on the feature matches shown in (a) are shown in (b) and (c), respectively. [sent-307, score-0.307]

45 In (d) the matched feature points after match growing are marked as red points. [sent-308, score-0.293]

46 The warps estimated by feature-based and fusion-based approaches based on the feature matches are shown in (e) and (f), respectively. [sent-309, score-0.307]

47 In the beginning, Tinit = T0, and then the temperature parameter T is gradually reduced Tnew = Told · rT, where rT ∈ (0, 1) is the annealing rate. [sent-310, score-0.298]

48 s F sumrtahlel amndor teh,e d correspondence set contains too many outliers, the initially estimated deformation parameters are far from the ground truth and the threshold d is also small. [sent-313, score-0.253]

49 Thus a large amount of points will be classified to be outliers which would include many inliers. [sent-314, score-0.27]

50 Therefore, in the beginning we set the maximum percentage of points that are allowed to be classified to outliers so that few inliers are falsely classified. [sent-315, score-0.485]

51 In the experiments, we show that with this bootstrap process, the recall of inliers is highly improved. [sent-317, score-0.316]

52 After all the outliers are removed, we obtain good estimates of the deformation parameters h via (11). [sent-320, score-0.353]

53 Fusion Optimization For regions lacking texture or having large deformations, the correspondences may not be well established and thus deformation parameters may not be well estimated by (11). [sent-323, score-0.284]

54 In this paper, we propose a fusion approach based on the TPS warp by exploiting both appearance and local features. [sent-324, score-0.319]

55 Match Growing As shown in the blue bounding box of Figure 2(a), the feature matches marked as red are concentrated in small regions which significantly degrades the registration accuracy as illustrated in Figure 2(b)-(c). [sent-360, score-0.4]

56 We propose a match growing approach based on the TPS model, which propagates the matches locally to establish more correspondences. [sent-362, score-0.338]

57 In the selected group, we use the matches to estimate the local TPS warp (9 control points are used) based on the feature-based approach. [sent-368, score-0.472]

58 Four matches estimated by W in each cell are added to the correspondence set. [sent-371, score-0.286]

59 In case that the propagated matches contain some outliers, 11999966 after obtaining the propagated matches, the proposed outlier rejection algorithm is used again to refine the matches. [sent-373, score-0.523]

60 Finally, we use the feature-based approach to estimate the TPS warp (100 control points are used) for the deformed instance, which is further refined by the fusion approach. [sent-374, score-0.466]

61 As shown in Figure 2(d) the matched feature points after match growing are uniformly distributed. [sent-375, score-0.293]

62 Thus, the estimated warp based on feature-based approach (Figure 2(e)) is better than that without growing (Figure 2(b)). [sent-376, score-0.28]

63 When the initial warp is far from the ground truth (Figure 2(b)), the refined warp obtained by the fusion approach is still not accurate (Figure 2(c)). [sent-377, score-0.466]

64 On the other hand, if the initial warp is close to the ground truth (Figure 2(e)), the fusion approach facilitates obtaining a more accurate warp (Figure 2(f)). [sent-378, score-0.466]

65 Experiments We present the detection and registration results on both synthetic and real images. [sent-380, score-0.26]

66 The outlier rejection is performed with T0= 10, Tfinal=2, rT=0. [sent-385, score-0.307]

67 For the quantitative evaluation of non-rigid surface detection and outlier rejection, the synthetic deformed surfaces are generated by TPS warp with 900 control points and these correspondences are considered as the ground truth. [sent-387, score-0.939]

68 We randomly sample 100 points in the template from the ground truth as the feature points denoted by Ft. [sent-395, score-0.322]

69 We further sample Nm points from Ft to construct the inlier matches for each instance. [sent-396, score-0.3]

70 Denote the number of instances by Ns and thus the total number of inlier matches is Nt = Ns Nm. [sent-397, score-0.289]

71 Let the ratio of outliers to inliers be ro and thus the ×nu Nmber of outliers is No = ro Nt. [sent-398, score-0.626]

72 To generate these outlier matches, we randomly sample Ft for No times, and the corresponding points in the input image are randomly sampled to construct spurious correspondences. [sent-399, score-0.284]

73 We compute the precision Pd and recall Rd of the surfaces to measure the detection performance. [sent-400, score-0.247]

74 Moreover, we also measure the precision Pm and recall Rm of the matches in the detected truth surfaces, which is critical for the registration task. [sent-401, score-0.495]

75 With the increase of outliers, the precision and recall of both detection and matching do not vary much. [sent-407, score-0.255]

76 However, the recall of matches decreases significantly when the number inliers is decreased (Figure 3(d)), which can be addressed by the proposed match growing algorithm. [sent-410, score-0.635]

77 In this experiment, we perform a comparative evaluation on the outlier rejection algorithms. [sent-412, score-0.307]

78 To evaluate the performance dependence on inliers, the number of inliers is varied from 20 to 110 with interval of 10. [sent-416, score-0.26]

79 The number of control points is set to be the same as the inliers but the locations are different. [sent-417, score-0.291]

80 In order to assess the robustness to the noise levels, the outlier correspondences are incrementally added. [sent-418, score-0.317]

81 The points in the template are randomly generated and the corresponding points in the synthetic images are also randomly generated. [sent-419, score-0.326]

82 We use the precision and recall of matches to measure the performance. [sent-424, score-0.316]

83 The precision and recall rates are monotonously increased with the increase of inliers, and the recall is always less than the precision. [sent-426, score-0.247]

84 The increases of precision and recall are similar when the number of inliers is small (less than 50). [sent-427, score-0.318]

85 However, when the number of inliers is further increased, the increase rate of recall is a little higher than precision. [sent-428, score-0.297]

86 The proposed algorithm is more robust to outliers when more inliers are obtained. [sent-431, score-0.388]

87 We compare our approach with the outlier rejection algorithm proposed in [24], LPOR (Local Topology based Outlier Rejection) for short, with the provided MATLAB implementation (http://isit. [sent-432, score-0.307]

88 Its performance decreases significantly with increase of outliers and its precision is always lower than recall. [sent-439, score-0.295]

89 to the fact that more outliers in the template will degrade the effect of the local topology structure assumed by the LPOR method. [sent-455, score-0.431]

90 On the contrary, the performance of our approach is only slightly decreased with increase of outlier level. [sent-456, score-0.252]

91 Even when the number of outliers is 19 times more than inlier, the precision is still above 0. [sent-457, score-0.26]

92 One example of matches before and after outlier rejection is shown in the Figure 4(e-f). [sent-464, score-0.489]

93 05 seconds to process the correspondence set with 110 inliers and 2200 outliers, and we refer it as fast AOR (FAOR for short). [sent-467, score-0.288]

94 For comparison, we use it for the outlier rejection problem by initializing its assignment matrix in accordance with the correspondences. [sent-477, score-0.307]

95 Our approach successfully detects all the deformed surfaces of the given templates and we also get good registration results. [sent-484, score-0.364]

96 Conclusions In this paper, we present an algorithm to detect and register all the nonrigid instances of given templates from noisy observations. [sent-489, score-0.298]

97 After obtaining the initial matches between the template and the input image, a novel high-order affinity graph is constructed to model the local topology consistency and then a hierarchical clustering approach is used to detect the nonrigid instances. [sent-490, score-0.739]

98 We propose a deterministic annealing approach based on the TPS model to remove the spurious matches in each cluster. [sent-491, score-0.546]

99 The proposed fusion approach exploits both appearance information and local features to register each nonrigid instance to the corresponding template. [sent-492, score-0.345]

100 To improve the registration performance, a TPS based match growing scheme is developed to propagate the matches. [sent-493, score-0.372]


similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('tps', 0.446), ('annealing', 0.207), ('outliers', 0.204), ('aor', 0.198), ('inliers', 0.184), ('warp', 0.183), ('matches', 0.182), ('outlier', 0.182), ('registration', 0.179), ('ctps', 0.163), ('template', 0.151), ('deformation', 0.149), ('ctp', 0.14), ('lpor', 0.14), ('nonrigid', 0.138), ('correspondences', 0.135), ('rejection', 0.125), ('affine', 0.111), ('qn', 0.108), ('correspondence', 0.104), ('surface', 0.1), ('fusion', 0.1), ('growing', 0.097), ('warps', 0.086), ('deterministic', 0.086), ('faor', 0.083), ('pn', 0.082), ('recall', 0.078), ('deformed', 0.076), ('topology', 0.076), ('surfaces', 0.075), ('register', 0.071), ('points', 0.066), ('zh', 0.066), ('match', 0.059), ('pilet', 0.057), ('gradually', 0.057), ('precision', 0.056), ('instances', 0.055), ('bootstrap', 0.054), ('inlier', 0.052), ('affinity', 0.05), ('locate', 0.05), ('ci', 0.049), ('hypergraph', 0.048), ('matching', 0.048), ('cho', 0.048), ('chertok', 0.047), ('tfinal', 0.047), ('tjm', 0.047), ('discrepancy', 0.046), ('warped', 0.045), ('ah', 0.044), ('fuses', 0.044), ('synthetic', 0.043), ('registering', 0.042), ('varied', 0.042), ('estimator', 0.041), ('ucmerced', 0.041), ('jz', 0.041), ('control', 0.041), ('feature', 0.039), ('graph', 0.039), ('omron', 0.038), ('pni', 0.038), ('plate', 0.038), ('clustering', 0.038), ('detection', 0.038), ('propagate', 0.037), ('deformable', 0.037), ('spurious', 0.036), ('triangulation', 0.036), ('zass', 0.036), ('appearance', 0.036), ('increase', 0.035), ('decreased', 0.035), ('transformation', 0.035), ('constructed', 0.035), ('remove', 0.035), ('bartoli', 0.034), ('temperature', 0.034), ('refine', 0.034), ('cells', 0.034), ('interval', 0.034), ('ratio', 0.034), ('sparsely', 0.034), ('thin', 0.034), ('templates', 0.034), ('rj', 0.033), ('merced', 0.033), ('matched', 0.032), ('residuals', 0.032), ('sz', 0.032), ('energy', 0.032), ('nm', 0.031), ('triangle', 0.031), ('beginning', 0.031), ('hierarchical', 0.03), ('ransac', 0.03), ('linearized', 0.03)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0000007 283 iccv-2013-Multiple Non-rigid Surface Detection and Registration

Author: Yi Wu, Yoshihisa Ijiri, Ming-Hsuan Yang

Abstract: Detecting and registering nonrigid surfaces are two important research problems for computer vision. Much work has been done with the assumption that there exists only one instance in the image. In this work, we propose an algorithm that detects and registers multiple nonrigid instances of given objects in a cluttered image. Specifically, after we use low level feature points to obtain the initial matches between templates and the input image, a novel high-order affinity graph is constructed to model the consistency of local topology. A hierarchical clustering approach is then used to locate the nonrigid surfaces. To remove the outliers in the cluster, we propose a deterministic annealing approach based on the Thin Plate Spline (TPS) model. The proposed method achieves high accuracy even when the number of outliers is nineteen times larger than the inliers. As the matches may appear sparsely in each instance, we propose a TPS based match growing approach to propagate the matches. Finally, an approach that fuses feature and appearance information is proposed to register each nonrigid surface. Extensive experiments and evaluations demonstrate that the proposed algorithm achieves promis- ing results in detecting and registering multiple non-rigid surfaces in a cluttered scene.

2 0.20157874 12 iccv-2013-A General Dense Image Matching Framework Combining Direct and Feature-Based Costs

Author: Jim Braux-Zin, Romain Dupont, Adrien Bartoli

Abstract: Dense motion field estimation (typically Romain Dupont1 romain . dupont @ cea . fr Adrien Bartoli2 adrien . bart o l @ gmai l com i . 2 ISIT, Universit e´ d’Auvergne/CNRS, France sions are explicitly modeled [32, 13]. Coarse-to-fine warping improves global convergence by making the assumption that optical flow, the motion of smaller structures is similar to the motion of stereo disparity and surface registration) is a key computer vision problem. Many solutions have been proposed to compute small or large displacements, narrow or wide baseline stereo disparity, but a unified methodology is still lacking. We here introduce a general framework that robustly combines direct and feature-based matching. The feature-based cost is built around a novel robust distance function that handles keypoints and “weak” features such as segments. It allows us to use putative feature matches which may contain mismatches to guide dense motion estimation out of local minima. Our framework uses a robust direct data term (AD-Census). It is implemented with a powerful second order Total Generalized Variation regularization with external and self-occlusion reasoning. Our framework achieves state of the art performance in several cases (standard optical flow benchmarks, wide-baseline stereo and non-rigid surface registration). Our framework has a modular design that customizes to specific application needs.

3 0.19137105 214 iccv-2013-Improving Graph Matching via Density Maximization

Author: Chao Wang, Lei Wang, Lingqiao Liu

Abstract: Graph matching has been widely used in various applications in computer vision due to its powerful performance. However, it poses three challenges to image sparse feature matching: (1) The combinatorial nature limits the size of the possible matches; (2) It is sensitive to outliers because the objective function prefers more matches; (3) It works poorly when handling many-to-many object correspondences, due to its assumption of one single cluster for each graph. In this paper, we address these problems with a unified framework—Density Maximization. We propose a graph density local estimator (퐷퐿퐸) to measure the quality of matches. Density Maximization aims to maximize the 퐷퐿퐸 values both locally and globally. The local maximization of 퐷퐿퐸 finds the clusters of nodes as well as eliminates the outliers. The global maximization of 퐷퐿퐸 efficiently refines the matches by exploring a much larger matching space. Our Density Maximization is orthogonal to specific graph matching algorithms. Experimental evaluation demonstrates that it significantly boosts the true matches and enables graph matching to handle both outliers and many-to-many object correspondences.

4 0.17039528 196 iccv-2013-Hierarchical Data-Driven Descent for Efficient Optimal Deformation Estimation

Author: Yuandong Tian, Srinivasa G. Narasimhan

Abstract: Real-world surfaces such as clothing, water and human body deform in complex ways. The image distortions observed are high-dimensional and non-linear, making it hard to estimate these deformations accurately. The recent datadriven descent approach [17] applies Nearest Neighbor estimators iteratively on a particular distribution of training samples to obtain a globally optimal and dense deformation field between a template and a distorted image. In this work, we develop a hierarchical structure for the Nearest Neighbor estimators, each of which can have only a local image support. We demonstrate in both theory and practice that this algorithm has several advantages over the nonhierarchical version: it guarantees global optimality with significantly fewer training samples, is several orders faster, provides a metric to decide whether a given image is “hard” (or “easy ”) requiring more (or less) samples, and can handle more complex scenes that include both global motion and local deformation. The proposed algorithm successfully tracks a broad range of non-rigid scenes including water, clothing, and medical images, and compares favorably against several other deformation estimation and tracking approaches that do not provide optimality guarantees.

5 0.15680164 16 iccv-2013-A Generic Deformation Model for Dense Non-rigid Surface Registration: A Higher-Order MRF-Based Approach

Author: Yun Zeng, Chaohui Wang, Xianfeng Gu, Dimitris Samaras, Nikos Paragios

Abstract: We propose a novel approach for dense non-rigid 3D surface registration, which brings together Riemannian geometry and graphical models. To this end, we first introduce a generic deformation model, called Canonical Distortion Coefficients (CDCs), by characterizing the deformation of every point on a surface using the distortions along its two principle directions. This model subsumes the deformation groups commonly used in surface registration such as isometry and conformality, and is able to handle more complex deformations. We also derive its discrete counterpart which can be computed very efficiently in a closed form. Based on these, we introduce a higher-order Markov Random Field (MRF) model which seamlessly integrates our deformation model and a geometry/texture similarity metric. Then we jointly establish the optimal correspondences for all the points via maximum a posteriori (MAP) inference. Moreover, we develop a parallel optimization algorithm to efficiently perform the inference for the proposed higher-order MRF model. The resulting registration algorithm outperforms state-of-the-art methods in both dense non-rigid 3D surface registration and tracking.

6 0.15517408 36 iccv-2013-Accurate and Robust 3D Facial Capture Using a Single RGBD Camera

7 0.15427649 358 iccv-2013-Robust Non-parametric Data Fitting for Correspondence Modeling

8 0.13548332 56 iccv-2013-Automatic Registration of RGB-D Scans via Salient Directions

9 0.13293046 432 iccv-2013-Uncertainty-Driven Efficiently-Sampled Sparse Graphical Models for Concurrent Tumor Segmentation and Atlas Registration

10 0.12560107 131 iccv-2013-EVSAC: Accelerating Hypotheses Generation by Modeling Matching Scores with Extreme Value Theory

11 0.12243074 27 iccv-2013-A Robust Analytical Solution to Isometric Shape-from-Template with Focal Length Calibration

12 0.1103427 140 iccv-2013-Elastic Net Constraints for Shape Matching

13 0.10662812 319 iccv-2013-Point-Based 3D Reconstruction of Thin Objects

14 0.10636665 121 iccv-2013-Discriminatively Trained Templates for 3D Object Detection: A Real Time Scalable Approach

15 0.1043692 270 iccv-2013-Modeling Self-Occlusions in Dynamic Shape and Appearance Tracking

16 0.10397162 238 iccv-2013-Learning Graphs to Match

17 0.10152102 183 iccv-2013-Geometric Registration Based on Distortion Estimation

18 0.10014099 139 iccv-2013-Elastic Fragments for Dense Scene Reconstruction

19 0.097921982 256 iccv-2013-Locally Affine Sparse-to-Dense Matching for Motion and Occlusion Estimation

20 0.096840382 11 iccv-2013-A Fully Hierarchical Approach for Finding Correspondences in Non-rigid Shapes


similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.207), (1, -0.12), (2, -0.047), (3, -0.004), (4, -0.019), (5, 0.02), (6, 0.018), (7, -0.022), (8, 0.033), (9, -0.029), (10, -0.041), (11, 0.075), (12, 0.021), (13, 0.081), (14, 0.106), (15, 0.047), (16, 0.126), (17, 0.133), (18, 0.103), (19, -0.08), (20, 0.066), (21, 0.105), (22, -0.045), (23, 0.037), (24, 0.085), (25, -0.105), (26, -0.016), (27, 0.193), (28, -0.047), (29, -0.037), (30, 0.094), (31, -0.083), (32, 0.153), (33, 0.024), (34, 0.071), (35, -0.071), (36, 0.001), (37, 0.014), (38, 0.056), (39, -0.001), (40, 0.052), (41, 0.117), (42, -0.018), (43, -0.03), (44, 0.006), (45, -0.001), (46, 0.05), (47, 0.033), (48, -0.013), (49, 0.002)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.95033646 283 iccv-2013-Multiple Non-rigid Surface Detection and Registration

Author: Yi Wu, Yoshihisa Ijiri, Ming-Hsuan Yang

Abstract: Detecting and registering nonrigid surfaces are two important research problems for computer vision. Much work has been done with the assumption that there exists only one instance in the image. In this work, we propose an algorithm that detects and registers multiple nonrigid instances of given objects in a cluttered image. Specifically, after we use low level feature points to obtain the initial matches between templates and the input image, a novel high-order affinity graph is constructed to model the consistency of local topology. A hierarchical clustering approach is then used to locate the nonrigid surfaces. To remove the outliers in the cluster, we propose a deterministic annealing approach based on the Thin Plate Spline (TPS) model. The proposed method achieves high accuracy even when the number of outliers is nineteen times larger than the inliers. As the matches may appear sparsely in each instance, we propose a TPS based match growing approach to propagate the matches. Finally, an approach that fuses feature and appearance information is proposed to register each nonrigid surface. Extensive experiments and evaluations demonstrate that the proposed algorithm achieves promis- ing results in detecting and registering multiple non-rigid surfaces in a cluttered scene.

2 0.77504706 16 iccv-2013-A Generic Deformation Model for Dense Non-rigid Surface Registration: A Higher-Order MRF-Based Approach

Author: Yun Zeng, Chaohui Wang, Xianfeng Gu, Dimitris Samaras, Nikos Paragios

Abstract: We propose a novel approach for dense non-rigid 3D surface registration, which brings together Riemannian geometry and graphical models. To this end, we first introduce a generic deformation model, called Canonical Distortion Coefficients (CDCs), by characterizing the deformation of every point on a surface using the distortions along its two principle directions. This model subsumes the deformation groups commonly used in surface registration such as isometry and conformality, and is able to handle more complex deformations. We also derive its discrete counterpart which can be computed very efficiently in a closed form. Based on these, we introduce a higher-order Markov Random Field (MRF) model which seamlessly integrates our deformation model and a geometry/texture similarity metric. Then we jointly establish the optimal correspondences for all the points via maximum a posteriori (MAP) inference. Moreover, we develop a parallel optimization algorithm to efficiently perform the inference for the proposed higher-order MRF model. The resulting registration algorithm outperforms state-of-the-art methods in both dense non-rigid 3D surface registration and tracking.

3 0.70849329 183 iccv-2013-Geometric Registration Based on Distortion Estimation

Author: Wei Zeng, Mayank Goswami, Feng Luo, Xianfeng Gu

Abstract: Surface registration plays a fundamental role in many applications in computer vision and aims at finding a oneto-one correspondence between surfaces. Conformal mapping based surface registration methods conformally map 2D/3D surfaces onto 2D canonical domains and perform the matching on the 2D plane. This registration framework reduces dimensionality, and the result is intrinsic to Riemannian metric and invariant under isometric deformation. However, conformal mapping will be affected by inconsistent boundaries and non-isometric deformations of surfaces. In this work, we quantify the effects of boundary variation and non-isometric deformation to conformal mappings, and give the theoretical upper bounds for the distortions of conformal mappings under these two factors. Besides giving the thorough theoretical proofs of the theorems, we verified them by concrete experiments using 3D human facial scans with dynamic expressions and varying boundaries. Furthermore, we used the distortion estimates for reducing search range in feature matching of surface registration applications. The experimental results are consistent with the theoreticalpredictions and also demonstrate the performance improvements in feature tracking.

4 0.69880801 185 iccv-2013-Go-ICP: Solving 3D Registration Efficiently and Globally Optimally

Author: Jiaolong Yang, Hongdong Li, Yunde Jia

Abstract: Registration is a fundamental task in computer vision. The Iterative Closest Point (ICP) algorithm is one of the widely-used methods for solving the registration problem. Based on local iteration, ICP is however well-known to suffer from local minima. Its performance critically relies on the quality of initialization, and only local optimality is guaranteed. This paper provides the very first globally optimal solution to Euclidean registration of two 3D pointsets or two 3D surfaces under the L2 error. Our method is built upon ICP, but combines it with a branch-and-bound (BnB) scheme which searches the 3D motion space SE(3) efficiently. By exploiting the special structure of the underlying geometry, we derive novel upper and lower bounds for the ICP error function. The integration of local ICP and global BnB enables the new method to run efficiently in practice, and its optimality is exactly guaranteed. We also discuss extensions, addressing the issue of outlier robustness.

5 0.6898123 196 iccv-2013-Hierarchical Data-Driven Descent for Efficient Optimal Deformation Estimation

Author: Yuandong Tian, Srinivasa G. Narasimhan

Abstract: Real-world surfaces such as clothing, water and human body deform in complex ways. The image distortions observed are high-dimensional and non-linear, making it hard to estimate these deformations accurately. The recent datadriven descent approach [17] applies Nearest Neighbor estimators iteratively on a particular distribution of training samples to obtain a globally optimal and dense deformation field between a template and a distorted image. In this work, we develop a hierarchical structure for the Nearest Neighbor estimators, each of which can have only a local image support. We demonstrate in both theory and practice that this algorithm has several advantages over the nonhierarchical version: it guarantees global optimality with significantly fewer training samples, is several orders faster, provides a metric to decide whether a given image is “hard” (or “easy ”) requiring more (or less) samples, and can handle more complex scenes that include both global motion and local deformation. The proposed algorithm successfully tracks a broad range of non-rigid scenes including water, clothing, and medical images, and compares favorably against several other deformation estimation and tracking approaches that do not provide optimality guarantees.

6 0.65478307 56 iccv-2013-Automatic Registration of RGB-D Scans via Salient Directions

7 0.65056819 140 iccv-2013-Elastic Net Constraints for Shape Matching

8 0.64210939 131 iccv-2013-EVSAC: Accelerating Hypotheses Generation by Modeling Matching Scores with Extreme Value Theory

9 0.63448548 358 iccv-2013-Robust Non-parametric Data Fitting for Correspondence Modeling

10 0.6140238 11 iccv-2013-A Fully Hierarchical Approach for Finding Correspondences in Non-rigid Shapes

11 0.59442258 139 iccv-2013-Elastic Fragments for Dense Scene Reconstruction

12 0.57320988 432 iccv-2013-Uncertainty-Driven Efficiently-Sampled Sparse Graphical Models for Concurrent Tumor Segmentation and Atlas Registration

13 0.56924695 214 iccv-2013-Improving Graph Matching via Density Maximization

14 0.54429585 224 iccv-2013-Joint Optimization for Consistent Multiple Graph Matching

15 0.53570712 205 iccv-2013-Human Re-identification by Matching Compositional Template with Cluster Sampling

16 0.52662146 12 iccv-2013-A General Dense Image Matching Framework Combining Direct and Feature-Based Costs

17 0.51516753 255 iccv-2013-Local Signal Equalization for Correspondence Matching

18 0.51338249 313 iccv-2013-Person Re-identification by Salience Matching

19 0.50861347 27 iccv-2013-A Robust Analytical Solution to Isometric Shape-from-Template with Focal Length Calibration

20 0.5054189 237 iccv-2013-Learning Graph Matching: Oriented to Category Modeling from Cluttered Scenes


similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(2, 0.055), (6, 0.011), (26, 0.068), (31, 0.027), (42, 0.068), (64, 0.033), (73, 0.413), (89, 0.218)]

similar papers list:

simIndex simValue paperId paperTitle

1 0.90397561 394 iccv-2013-Single-Patch Low-Rank Prior for Non-pointwise Impulse Noise Removal

Author: Ruixuan Wang, Emanuele Trucco

Abstract: This paper introduces a ‘low-rank prior’ for small oriented noise-free image patches: considering an oriented patch as a matrix, a low-rank matrix approximation is enough to preserve the texture details in the properly oriented patch. Based on this prior, we propose a single-patch method within a generalized joint low-rank and sparse matrix recovery framework to simultaneously detect and remove non-pointwise random-valued impulse noise (e.g., very small blobs). A weighting matrix is incorporated in the framework to encode an initial estimate of the spatial noise distribution. An accelerated proximal gradient method is adapted to estimate the optimal noise-free image patches. Experiments show the effectiveness of our framework in removing non-pointwise random-valued impulse noise.

2 0.8936525 98 iccv-2013-Cross-Field Joint Image Restoration via Scale Map

Author: Qiong Yan, Xiaoyong Shen, Li Xu, Shaojie Zhuo, Xiaopeng Zhang, Liang Shen, Jiaya Jia

Abstract: Color, infrared, and flash images captured in different fields can be employed to effectively eliminate noise and other visual artifacts. We propose a two-image restoration framework considering input images in different fields, for example, one noisy color image and one dark-flashed nearinfrared image. The major issue in such a framework is to handle structure divergence and find commonly usable edges and smooth transition for visually compelling image reconstruction. We introduce a scale map as a competent representation to explicitly model derivative-level confidence and propose new functions and a numerical solver to effectively infer it following new structural observations. Our method is general and shows a principled way for cross-field restoration.

3 0.84527963 58 iccv-2013-Bayesian 3D Tracking from Monocular Video

Author: Ernesto Brau, Jinyan Guan, Kyle Simek, Luca Del Pero, Colin Reimer Dawson, Kobus Barnard

Abstract: Jinyan Guan† j guan1 @ emai l ari z ona . edu . Kyle Simek† ks imek@ emai l ari z ona . edu . Colin Reimer Dawson‡ cdaws on@ emai l ari z ona . edu . ‡School of Information University of Arizona Kobus Barnard‡ kobus @ s i sta . ari z ona . edu ∗School of Informatics University of Edinburgh for tracking an unknown and changing number of people in a scene using video taken from a single, fixed viewpoint. We develop a Bayesian modeling approach for tracking people in 3D from monocular video with unknown cameras. Modeling in 3D provides natural explanations for occlusions and smoothness discontinuities that result from projection, and allows priors on velocity and smoothness to be grounded in physical quantities: meters and seconds vs. pixels and frames. We pose the problem in the context of data association, in which observations are assigned to tracks. A correct application of Bayesian inference to multitarget tracking must address the fact that the model’s dimension changes as tracks are added or removed, and thus, posterior densities of different hypotheses are not comparable. We address this by marginalizing out the trajectory parameters so the resulting posterior over data associations has constant dimension. This is made tractable by using (a) Gaussian process priors for smooth trajectories and (b) approximately Gaussian likelihood functions. Our approach provides a principled method for incorporating multiple sources of evidence; we present results using both optical flow and object detector outputs. Results are comparable to recent work on 3D tracking and, unlike others, our method requires no pre-calibrated cameras.

same-paper 4 0.83668578 283 iccv-2013-Multiple Non-rigid Surface Detection and Registration

Author: Yi Wu, Yoshihisa Ijiri, Ming-Hsuan Yang

Abstract: Detecting and registering nonrigid surfaces are two important research problems for computer vision. Much work has been done with the assumption that there exists only one instance in the image. In this work, we propose an algorithm that detects and registers multiple nonrigid instances of given objects in a cluttered image. Specifically, after we use low level feature points to obtain the initial matches between templates and the input image, a novel high-order affinity graph is constructed to model the consistency of local topology. A hierarchical clustering approach is then used to locate the nonrigid surfaces. To remove the outliers in the cluster, we propose a deterministic annealing approach based on the Thin Plate Spline (TPS) model. The proposed method achieves high accuracy even when the number of outliers is nineteen times larger than the inliers. As the matches may appear sparsely in each instance, we propose a TPS based match growing approach to propagate the matches. Finally, an approach that fuses feature and appearance information is proposed to register each nonrigid surface. Extensive experiments and evaluations demonstrate that the proposed algorithm achieves promis- ing results in detecting and registering multiple non-rigid surfaces in a cluttered scene.

5 0.81749117 12 iccv-2013-A General Dense Image Matching Framework Combining Direct and Feature-Based Costs

Author: Jim Braux-Zin, Romain Dupont, Adrien Bartoli

Abstract: Dense motion field estimation (typically Romain Dupont1 romain . dupont @ cea . fr Adrien Bartoli2 adrien . bart o l @ gmai l com i . 2 ISIT, Universit e´ d’Auvergne/CNRS, France sions are explicitly modeled [32, 13]. Coarse-to-fine warping improves global convergence by making the assumption that optical flow, the motion of smaller structures is similar to the motion of stereo disparity and surface registration) is a key computer vision problem. Many solutions have been proposed to compute small or large displacements, narrow or wide baseline stereo disparity, but a unified methodology is still lacking. We here introduce a general framework that robustly combines direct and feature-based matching. The feature-based cost is built around a novel robust distance function that handles keypoints and “weak” features such as segments. It allows us to use putative feature matches which may contain mismatches to guide dense motion estimation out of local minima. Our framework uses a robust direct data term (AD-Census). It is implemented with a powerful second order Total Generalized Variation regularization with external and self-occlusion reasoning. Our framework achieves state of the art performance in several cases (standard optical flow benchmarks, wide-baseline stereo and non-rigid surface registration). Our framework has a modular design that customizes to specific application needs.

6 0.69035095 399 iccv-2013-Spoken Attributes: Mixing Binary and Relative Attributes to Say the Right Thing

7 0.68885791 223 iccv-2013-Joint Noise Level Estimation from Personal Photo Collections

8 0.68187428 60 iccv-2013-Bayesian Robust Matrix Factorization for Image and Video Processing

9 0.68022984 358 iccv-2013-Robust Non-parametric Data Fitting for Correspondence Modeling

10 0.67340785 23 iccv-2013-A New Image Quality Metric for Image Auto-denoising

11 0.63701785 196 iccv-2013-Hierarchical Data-Driven Descent for Efficient Optimal Deformation Estimation

12 0.63569403 27 iccv-2013-A Robust Analytical Solution to Isometric Shape-from-Template with Focal Length Calibration

13 0.63282704 151 iccv-2013-Exploiting Reflection Change for Automatic Reflection Removal

14 0.63049573 300 iccv-2013-Optical Flow via Locally Adaptive Fusion of Complementary Data Costs

15 0.6270858 105 iccv-2013-DeepFlow: Large Displacement Optical Flow with Deep Matching

16 0.62121069 304 iccv-2013-PM-Huber: PatchMatch with Huber Regularization for Stereo Matching

17 0.61982334 432 iccv-2013-Uncertainty-Driven Efficiently-Sampled Sparse Graphical Models for Concurrent Tumor Segmentation and Atlas Registration

18 0.61950088 351 iccv-2013-Restoring an Image Taken through a Window Covered with Dirt or Rain

19 0.61838681 230 iccv-2013-Latent Data Association: Bayesian Model Selection for Multi-target Tracking

20 0.61675137 288 iccv-2013-Nested Shape Descriptors