cvpr cvpr2013 cvpr2013-240 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Samuele Salti, Alessandro Lanza, Luigi Di_Stefano
Abstract: The paper conjectures and demonstrates that repeatable keypoints based on salient symmetries at different scales can be detected by a novel analysis grounded on the wave equation rather than the heat equation underlying traditional Gaussian scale–space theory. While the image structures found by most state-of-the-art detectors, such as blobs and corners, occur typically on planar highly textured surfaces, salient symmetries are widespread in diverse kinds of images, including those related to untextured objects, which are hardly dealt with by current feature-based recognition pipelines. We provide experimental results on standard datasets and also contribute with a new dataset focused on untextured objects. Based on the positive experimental results, we hope to foster further research on the promising topic ofscale invariant analysis through the wave equation.
Reference: text
sentIndex sentText sentNum sentScore
1 it Abstract The paper conjectures and demonstrates that repeatable keypoints based on salient symmetries at different scales can be detected by a novel analysis grounded on the wave equation rather than the heat equation underlying traditional Gaussian scale–space theory. [sent-7, score-1.433]
2 We provide experimental results on standard datasets and also contribute with a new dataset focused on untextured objects. [sent-9, score-0.208]
3 Based on the positive experimental results, we hope to foster further research on the promising topic ofscale invariant analysis through the wave equation. [sent-10, score-0.608]
4 The most popular is certainly the heat or diffusion equation, but other non-linear second-order [16], fourthorder [12] and, very recently, fractional-order PDEs [3] have been deployed for image processing. [sent-15, score-0.159]
5 Although the great majority of these equations are parabolic, researchers are also investigating on the use of hyperbolic equations, such as the shock-filters [15] or the telegrapher equation [17]. [sent-16, score-0.133]
6 The scale–space theory was developed in seminal works by Witkin [24] and Koenderink (a) (b) (c) (d) Figure 1: (a-b) keypoints by our algorithm; (c-d) keypoints by DoG [10]. [sent-18, score-0.436]
7 Symmetries are more likely than blobs to appear in untextured objects (a-c): only a few DoG keypoints actually lay on the object. [sent-19, score-0.453]
8 On the other hand, a similar number of keypoints is detected on textured surfaces (b-d): yet, the proposed keypoints concentrate on more evident (i. [sent-20, score-0.514]
9 Instead, in work dating back to the 90’s [6, 4, 19], the wave equation was used for skeletonization of binary silhouettes and detection of circular objects, due to its ability of eliciting the symmetry set. [sent-26, score-0.831]
10 More recently, PDEs other than the heat equation have been used for description of global silhouettes [21]. [sent-27, score-0.148]
11 In particular, we focus on the problem of keypoints detection and advocate the use of the wave equation. [sent-29, score-0.802]
12 When applied to images, the wave equation exhibits the ability to highlight symmetries at different scales. [sent-30, score-0.861]
13 Salient symmetries are likely to be found in a large range of images, and in particular in those related to man-made untextured objects (Fig. [sent-31, score-0.423]
14 Nevertheless, many salient symmetries arise in textured objects alike (Fig. [sent-33, score-0.309]
15 Thus, the proposed approach qualifies as a general tool for repeatable salient regions detection. [sent-35, score-0.18]
16 Surprisingly, there has been relatively little work on the use of symmetry as a cue to detect and describe local features. [sent-36, score-0.153]
17 The most recent contribution on this topic is due to Haugge et Snavely [7], who propose a detector-descriptor pair relying on a local symmetry score computed densely on the image and across scales. [sent-37, score-0.123]
18 Unlike our proposal, though, their formalization of symmetry is specifically engineered to capture effectively the salient regions likely to be found in architectural images and relies on the classic Gaussian scale–space rather than our novel formulation grounded on the wave PDE. [sent-38, score-0.749]
19 Earlier related works such as [18, 11] focus on detecting interest points featuring radial symmetries (e. [sent-39, score-0.215]
20 The wave equation The wave equation is a well-known linear second-order hyperbolic partial differential equation. [sent-43, score-1.334]
21 The boundary conditions, defined by the fourth equation, represent the first order highly absorbing local approximation of the perfectly absorbing boundary conditions [5], which are inherently non-local in space and time, i. [sent-50, score-0.255]
22 Absorbing boundary conditions have been chosen to avoid reflections of waves at image boundaries, that would cause unwanted interferences. [sent-53, score-0.162]
23 , w, with w and h the image width and height in pixels, respectively), denoting by r = cΔt/Δx ∈ R+ the Courant nspuemcbtivere layn)d, substituting trhe = =ab cΔovet/ fΔinxite ∈ ∈di Rfference formulae into the wave equation, we obtain the following explicit time-marching scheme: ×× uin,+j1=r42? [sent-72, score-0.615]
24 ti,own −at 1th}e × remaining hn −od 1e}s, we exploit t Theo initial conditions, whose discretization easily yields an explicit solution for the inner nodes at the first iteration: − − ui1,j=r82? [sent-92, score-0.095]
25 discretization yields an explicit solution for the image boundary nodes: uin,+11 = uin,1 + r(uin,2 − uin,1) u i1n , w+ j1 = u i1n , wj+ r ( u 2nin, wj− 1 u−1n,j u)in,w) uhn+,j1 = uhn,j + r(uhn−1,j − uhn,j) 222888999977 . [sent-96, score-0.113]
26 (c) Figure 2: (a) initial image I; (b-c) function u(x, y; t), from a side and a top view respectively, at the time t when the wave has traveled a distance equal to the radius of the circle (graylevel intensities get higher from blue to red). [sent-114, score-0.782]
27 the space traveled by the wave in one unit of discretized time, rather than the two constants individually. [sent-117, score-0.674]
28 obtain this by setting c = Despite this constraint on Δt, we can use an explicit scheme because we are not interested in computing with few iterations the solution at large times, but rather in analyzing the wave evolution with time steps commensurable to the image lattice. [sent-122, score-0.673]
29 Wave–based scale analysis The evolution of image intensities obtained through the wave equation (1) allows for multi–scale signal analysis. [sent-125, score-0.764]
30 In particular, as shown in [6], by simulating wave propagation on images it is possible to detect circles of varying radii. [sent-126, score-0.68]
31 2a, and think of it as a wave at time 0, then the wave front after the w? [sent-128, score-1.168]
32 e has traveled a distance equal to the radius of the circle is given by the? [sent-133, score-0.173]
33 sum of circular waves originating from each point on the circle edge. [sent-137, score-0.155]
34 Therefore, the wave propagation attains an extremum at the center of the circle at a time that is proportional to the scale of the original circle. [sent-138, score-0.926]
35 Detection of circle centers is actually a particular case of the more general property of wave propagation of eliciting the symmetry set of curves [19]. [sent-141, score-0.887]
36 The symmetry set is defined as the locus of centers of circles bi-tangent to a curve and can be detected by summing u(x, y; t) over t into an accumulator and inserting at each iteration the local spatial extrema of the accumulator into the set. [sent-142, score-0.513]
37 To define repeatable keypoints, only the points of locally maximal symmetry must be considered, which correspond to the shock ? [sent-144, score-0.306]
38 Figure 3: Temporal evolution of the intensity at the center of the circle in Fig. [sent-162, score-0.113]
39 2a under the discretized wave process (left) and the discretized wave-diffusion process (right). [sent-163, score-0.668]
40 points of the wave propagation (see [21] for a deep analysis of shock points). [sent-164, score-0.694]
41 Indeed, one interesting future work is the theoretical investigation on the definition of a sound scale–space theory from the wave equation. [sent-167, score-0.61]
42 Unfortunately, discretization of the wave propagation is prone to quite significant numerical errors, mainly due to numerical (or grid) dispersion [23]. [sent-168, score-0.903]
43 To overcome such issues, we adopt a solution similar to that proposed in [19]: we interleave a wave prop- agation step to a linear diffusion step governed by the heat equation with diffusivity k ∈ R+, ut(x, y; t) = k∇2u? [sent-171, score-0.812]
44 images: in the considered example, three extrema show up, two corresponding to the symmetries arising at the center of the square-like structures formed by the sides of the rectangle and one corresponding to the center of the rectangle. [sent-227, score-0.478]
45 The time of an extremum (xext, yext, text) is related to the scale (the radius) rext of the detected symmetry by the following simple relation: rext=cΔtexxt=cΔΔtxnext. [sent-228, score-0.444]
46 Given such scale covariant extrema in our family of signals, we define as keypoints the sharp local extrema of u(x, y; t). [sent-231, score-0.871]
47 Although on the synthetic images showed so far the requirement of sharpness of local extrema is redundant, as all extrema are indeed sharp, in real images, weak (i. [sent-232, score-0.578]
48 not repeatable) symmetries may be detected if all local extrema are accepted. [sent-234, score-0.504]
49 The extrema (minima) show up at times proportional to the circles radius. [sent-236, score-0.294]
50 We define a sharp extremum as one whose value is significantly larger (or smaller) than the average of the values in its temporal neighborhood: |u(xext, yext, text) −¯ u(xext, yext, text)| ≥ θ, (10) u¯(xext,yext,text) =t2−1 t1k? [sent-238, score-0.17]
51 (11) Both the neighborhood [t1, t2] and the threshold θ must adapt with scale, as the absolute value at the extrema as well as their smoothness increase with scale (Fig. [sent-240, score-0.374]
52 However, spatial neighborhoods are not defined at the edge of the image and the computation of the average is computationally more demanding than over 1-D temporal neighborhoods (efficient schemes, such as integral images or histograms can be, of course, deployed in both cases). [sent-244, score-0.096]
53 To limit the influence of clutter and occlusions, we select as end of the neighborhood the extremum itself, t2 = text, so that any texture external to the symmetry does not influence its detection. [sent-248, score-0.351]
54 The trend of u(x, y; t) at the center of the circle starts with a plateau and then exhibits a sudden drop toward the extremum: indeed, this would be a discontinuity in the continuous case. [sent-253, score-0.119]
55 Therefore, we define the starting point of the neighborhood as the end of the plateau (Fig. [sent-254, score-0.13]
56 As the trend deviates more and more from the continuous ideal solution when the scale gets larger, because of the numerical dispersion and the diffusion that we interleave with propagation, we learn a starting point for each scale by regressing the neighborhood of the circle (Fig. [sent-256, score-0.475]
57 (13) As we are interested in detecting weaker, though repeatable, symmetries than a high-contrast circle, the actual value θ used at each scale is defined as a fraction ρ of θmax (rext). [sent-266, score-0.275]
58 The domain for searching local extrema must also take into account the peculiarity ofthe discretized wave propagation. [sent-267, score-0.889]
59 The most straightforward estimation of local extrema of 3D continuous signals, given their discretized version, is represented by extrema within a 3x3x3 neighborhood in the discrete domain. [sent-268, score-0.619]
60 However, in wave propagation time and space discretizations are bound by the constant c, which means, in particular, that it takes cΔ1t discrete time intervals for the wave to travel 1 pixel. [sent-269, score-1.233]
61 Therefore, to correctly determine if the wave state at (x, y, t) is a local extremum we have to compare it not only with its 3x3 neighbors at time t − 1 and t + 1, but with all the 3x3 neighbors at time tki ∈e t[t − −− 1c aΔn1td, tt ++ 1c,Δ1 btu]. [sent-270, score-0.721]
62 t Iint hot ahlle trh weo 3rxd3s, n teiog rhebaodr sth aet corrkec ∈t wave state in past and future times we have to give to all the contributions the time to move to a given pixel from its neighbors, process that can take at most cΔ1t time steps. [sent-271, score-0.584]
63 With our choice of Δt and c, we have cΔ1t and, hence, it turns out that we have to perform the search for local extrema in a 3x3x5 neighborhood. [sent-272, score-0.263]
64 It includes 8 planar scenes and 5 nuisance factors: scale and rotation changes, viewpoint changes, decreasing illumination, blur and JPEG compression. [sent-279, score-0.129]
65 Performance is measured according to repeatability and number of correct correspondences. [sent-280, score-0.115]
66 Each scene is captured from a set of repeatable positions along four paths by using a robotic arm. [sent-283, score-0.138]
67 The authors consider only the recall rate, an analogous of repeatability in the previous dataset. [sent-285, score-0.115]
68 Moreover, to test the performance of detectors on challenging, untextured objects, we introduce a novel dataset consisting of 5 man-made objects (Fig. [sent-286, score-0.255]
69 Such implementation, together with our untextured objects dataset are publicly available at the project website1 . [sent-292, score-0.208]
70 (5) 6: end for 7: for all boundary pixels (i, j) do 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23: 24: ui21,j ← eq. [sent-310, score-0.092]
71 (8) end for for all boundary pixels (i, j) do ui1,j ← eq. [sent-312, score-0.092]
72 (4) end for for all boundary pixels (i, j) do uin,+j21 ← eq. [sent-316, score-0.092]
73 (8) 26: end for 27: for all boundary pixels (i, j) do 28: uin,+j1 ← eq. [sent-319, score-0.092]
74 We start to search for extrema after 12 iterations (i. [sent-322, score-0.263]
75 We present repeatability charts for each of the nuisances of the dataset. [sent-332, score-0.174]
76 Moreover, the intuition that symmetries are a powerful and robust cue for finding repeatable keypoints is supported by the large margin in performance between our proposal and previous detectors based on other cues. [sent-335, score-0.688]
77 7c), which deal with robustness to affine deformations, our detector provides the same performance as DoG: one major drawback of both detectors is the lack of an affine renormalization of the extracted patch, which hinders their results in such comparison. [sent-339, score-0.131]
78 7e) show how symmetries are definitely the best cue to overcome blur: this makes sense as, intuitively, symmetries are largely unaffected by isotropic smoothing. [sent-343, score-0.46]
79 Hence results for our method are available only at the minimum, middle and maximum scale variation, whereas results for all other detectors were kindly provided by the authors of [1]. [sent-354, score-0.107]
80 The proposed algorithm shows the best recall rate at the middle scale variation and outperforms by a large margin all the other detectors at the maximum scale variation. [sent-355, score-0.167]
81 g-h) Mean repeatability and number of correspondences on the 5 sets of images of the Untextured dataset. [sent-365, score-0.14]
82 Figure 8: Correct matches of SIFT descriptors versus overlap error of keypoints on Oxford and untextured datasets. [sent-368, score-0.426]
83 7g-7h report the results on our novel, untextured objects dataset. [sent-372, score-0.208]
84 We provide the mean repeatability and mean number of correspondences for each method. [sent-373, score-0.14]
85 7h) by two orders of magnitude with respect to the other datasets, and repeatability falls below 50%. [sent-376, score-0.115]
86 Please note that on this dataset we lower from 40% to 20% the overlap error to consider two keypoints as repeatable: as shown by Fig. [sent-377, score-0.218]
87 SIFT) can correctly match about 51% of the keypoints whose overlap error is smaller than 40% but on our tougher dataset only 38% of such keypoints can be matched, and 48% can be achieved only with overlap errors smaller than 10%. [sent-380, score-0.436]
88 Therefore, we found that with untextured objects keypoints should be localized more accurately, when used with existing descriptors, to reach performance comparable to those attained on planar, highly textured objects. [sent-381, score-0.478]
89 Symmetries confirm to be a robust cue even in this challenging scenario: their repeatability is the least affected by the increasing difficulties of the scenes. [sent-383, score-0.145]
90 They also turn out to provide the highest number of correspondences, an important practical trait to enable their use in recognition of untextured objects in presence of occlusions. [sent-384, score-0.208]
91 Instead of the heat equation, hyperbolic PDEs, and in par222999000422 ticular the wave equation, can indeed be used to derive a one-parameter family of signals that enables scale–invariant image analysis. [sent-387, score-0.786]
92 The wave equation fires on symmetries, and we have shown these to provide robust and discriminative cues for detection of repeatable keypoints on a broader set of image structures than those found on planar highly textured surfaces. [sent-388, score-1.086]
93 As shown by prominent evaluations [1, 14], salient regions are usually complementary, so that a bunch of diverse detectors might be deployed whenever one cannot predict in advance the likely scene structures. [sent-389, score-0.127]
94 Next, more sophisticated numerical schemes for discretizing the wave equation for feature detection might be helpful: in particular, to reduce numerical dispersion and diffusion as well as to achieve higher rotational invariance. [sent-394, score-0.905]
95 Finally, several interesting extensions are possible, such as defining a symmetry descriptor from u(x, y; t) and extending feature detection to color images via the model of images as manifold embedded in a com- bined spatial-color space [20]. [sent-395, score-0.123]
96 Detection of circular objects by wave propagation on a mesh-connected computer. [sent-428, score-0.676]
97 Absorbing boundary conditions for the numerical simulation of waves. [sent-433, score-0.167]
98 Using a mixed [20] [21] [22] [23] [24] wave/diffusion process to elicit the symmetry set. [sent-548, score-0.123]
99 Symmetry maps of free-form curve segments via wave propagation. [sent-562, score-0.584]
100 On the numerical solution of the 2D wave equation with compact FDTD schemes. [sent-578, score-0.726]
wordName wordTfidf (topN-words)
[('wave', 0.584), ('uin', 0.337), ('extrema', 0.263), ('keypoints', 0.218), ('symmetries', 0.215), ('untextured', 0.208), ('repeatable', 0.138), ('extremum', 0.137), ('symmetry', 0.123), ('repeatability', 0.115), ('pdes', 0.115), ('rext', 0.098), ('oxford', 0.086), ('heat', 0.086), ('circle', 0.08), ('numerical', 0.08), ('unibo', 0.069), ('aug', 0.067), ('propagation', 0.065), ('dispersion', 0.064), ('equation', 0.062), ('scale', 0.06), ('graylevel', 0.059), ('wade', 0.059), ('xext', 0.059), ('yext', 0.059), ('absorbing', 0.058), ('dog', 0.054), ('bologna', 0.052), ('textured', 0.052), ('sharpness', 0.052), ('boundary', 0.052), ('neighborhood', 0.051), ('waves', 0.048), ('traveled', 0.048), ('detectors', 0.047), ('interleave', 0.045), ('shock', 0.045), ('radius', 0.045), ('discretized', 0.042), ('salient', 0.042), ('hyperbolic', 0.042), ('proposal', 0.04), ('end', 0.04), ('alt', 0.04), ('signals', 0.04), ('binaries', 0.039), ('lanza', 0.039), ('plateau', 0.039), ('rcm', 0.039), ('uhn', 0.039), ('deployed', 0.038), ('mar', 0.037), ('nuisance', 0.037), ('robot', 0.035), ('conditions', 0.035), ('rmax', 0.035), ('rmin', 0.035), ('eliciting', 0.035), ('luigi', 0.035), ('accumulator', 0.035), ('bikes', 0.035), ('ibr', 0.035), ('diffusion', 0.035), ('family', 0.034), ('inner', 0.034), ('evolution', 0.033), ('sharp', 0.033), ('ebr', 0.032), ('planar', 0.032), ('tn', 0.032), ('explicit', 0.031), ('circles', 0.031), ('cue', 0.03), ('mser', 0.03), ('charts', 0.03), ('haifa', 0.03), ('discretization', 0.03), ('nuisances', 0.029), ('neighborhoods', 0.029), ('equations', 0.029), ('boat', 0.028), ('detector', 0.028), ('affine', 0.028), ('tuytelaars', 0.027), ('unwanted', 0.027), ('blobs', 0.027), ('circular', 0.027), ('theoretical', 0.026), ('stefano', 0.026), ('jul', 0.026), ('detected', 0.026), ('correspondences', 0.025), ('nov', 0.025), ('scheme', 0.025), ('intensities', 0.025), ('mesh', 0.025), ('jpeg', 0.024), ('invariant', 0.024), ('yj', 0.024)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999928 240 cvpr-2013-Keypoints from Symmetries by Wave Propagation
Author: Samuele Salti, Alessandro Lanza, Luigi Di_Stefano
Abstract: The paper conjectures and demonstrates that repeatable keypoints based on salient symmetries at different scales can be detected by a novel analysis grounded on the wave equation rather than the heat equation underlying traditional Gaussian scale–space theory. While the image structures found by most state-of-the-art detectors, such as blobs and corners, occur typically on planar highly textured surfaces, salient symmetries are widespread in diverse kinds of images, including those related to untextured objects, which are hardly dealt with by current feature-based recognition pipelines. We provide experimental results on standard datasets and also contribute with a new dataset focused on untextured objects. Based on the positive experimental results, we hope to foster further research on the promising topic ofscale invariant analysis through the wave equation.
2 0.14785817 234 cvpr-2013-Joint Spectral Correspondence for Disparate Image Matching
Author: Mayank Bansal, Kostas Daniilidis
Abstract: We address the problem of matching images with disparate appearance arising from factors like dramatic illumination (day vs. night), age (historic vs. new) and rendering style differences. The lack of local intensity or gradient patterns in these images makes the application of pixellevel descriptors like SIFT infeasible. We propose a novel formulation for detecting and matching persistent features between such images by analyzing the eigen-spectrum of the joint image graph constructed from all the pixels in the two images. We show experimental results of our approach on a public dataset of challenging image pairs and demonstrate significant performance improvements over state-of-the-art.
Author: Stefan Harmeling, Michael Hirsch, Bernhard Schölkopf
Abstract: We establish a link between Fourier optics and a recent construction from the machine learning community termed the kernel mean map. Using the Fraunhofer approximation, it identifies the kernel with the squared Fourier transform of the aperture. This allows us to use results about the invertibility of the kernel mean map to provide a statement about the invertibility of Fraunhofer diffraction, showing that imaging processes with arbitrarily small apertures can in principle be invertible, i.e., do not lose information, provided the objects to be imaged satisfy a generic condition. A real world experiment shows that we can super-resolve beyond the Rayleigh limit.
4 0.086079702 432 cvpr-2013-Three-Dimensional Bilateral Symmetry Plane Estimation in the Phase Domain
Author: Ramakrishna Kakarala, Prabhu Kaliamoorthi, Vittal Premachandran
Abstract: We show that bilateral symmetry plane estimation for three-dimensional (3-D) shapes may be carried out accurately, and efficiently, in the spherical harmonic domain. Our methods are valuable for applications where spherical harmonic expansion is already employed, such as 3-D shape registration, morphometry, and retrieval. We show that the presence of bilateral symmetry in the 3-D shape is equivalent to a linear phase structure in the corresponding spherical harmonic coefficients, and provide algorithms for estimating the orientation of the symmetry plane. The benefit of using spherical harmonic phase is that symmetry estimation reduces to matching a compact set of descriptors, without the need to solve a correspondence problem. Our methods work on point clouds as well as large-scale mesh models of 3-D shapes.
5 0.069519125 192 cvpr-2013-Graph Matching with Anchor Nodes: A Learning Approach
Author: Nan Hu, Raif M. Rustamov, Leonidas Guibas
Abstract: In this paper, we consider the weighted graph matching problem with partially disclosed correspondences between a number of anchor nodes. Our construction exploits recently introduced node signatures based on graph Laplacians, namely the Laplacian family signature (LFS) on the nodes, and the pairwise heat kernel map on the edges. In this paper, without assuming an explicit form of parametric dependence nor a distance metric between node signatures, we formulate an optimization problem which incorporates the knowledge of anchor nodes. Solving this problem gives us an optimized proximity measure specific to the graphs under consideration. Using this as a first order compatibility term, we then set up an integer quadratic program (IQP) to solve for a near optimal graph matching. Our experiments demonstrate the superior performance of our approach on randomly generated graphs and on two widelyused image sequences, when compared with other existing signature and adjacency matrix based graph matching methods.
6 0.064954959 248 cvpr-2013-Learning Collections of Part Models for Object Recognition
7 0.054707538 303 cvpr-2013-Multi-view Photometric Stereo with Spatially Varying Isotropic Materials
8 0.054572724 298 cvpr-2013-Multi-scale Curve Detection on Surfaces
9 0.053424135 286 cvpr-2013-Mirror Surface Reconstruction from a Single Image
10 0.052532688 91 cvpr-2013-Consensus of k-NNs for Robust Neighborhood Selection on Graph-Based Manifolds
11 0.050995182 231 cvpr-2013-Joint Detection, Tracking and Mapping by Semantic Bundle Adjustment
12 0.050293002 208 cvpr-2013-Hyperbolic Harmonic Mapping for Constrained Brain Surface Registration
13 0.047668066 196 cvpr-2013-HON4D: Histogram of Oriented 4D Normals for Activity Recognition from Depth Sequences
15 0.046078976 449 cvpr-2013-Unnatural L0 Sparse Representation for Natural Image Deblurring
16 0.045330379 444 cvpr-2013-Unconstrained Monocular 3D Human Pose Estimation by Action Detection and Cross-Modality Regression Forest
17 0.044189136 325 cvpr-2013-Part Discovery from Partial Correspondence
18 0.043859068 438 cvpr-2013-Towards Pose Robust Face Recognition
19 0.04349592 137 cvpr-2013-Dynamic Scene Classification: Learning Motion Descriptors with Slow Features Analysis
20 0.043334477 104 cvpr-2013-Deep Convolutional Network Cascade for Facial Point Detection
topicId topicWeight
[(0, 0.125), (1, 0.037), (2, 0.023), (3, 0.024), (4, 0.008), (5, 0.002), (6, -0.013), (7, -0.02), (8, -0.009), (9, -0.022), (10, 0.0), (11, 0.004), (12, -0.007), (13, -0.051), (14, 0.018), (15, -0.049), (16, 0.027), (17, -0.007), (18, 0.06), (19, 0.042), (20, 0.001), (21, 0.017), (22, 0.015), (23, 0.033), (24, -0.001), (25, 0.0), (26, 0.034), (27, -0.042), (28, 0.008), (29, 0.039), (30, 0.029), (31, -0.001), (32, 0.021), (33, -0.0), (34, 0.043), (35, 0.012), (36, 0.076), (37, 0.029), (38, 0.006), (39, 0.02), (40, 0.041), (41, 0.041), (42, -0.052), (43, 0.067), (44, -0.023), (45, -0.005), (46, 0.019), (47, 0.024), (48, 0.057), (49, 0.02)]
simIndex simValue paperId paperTitle
same-paper 1 0.8984856 240 cvpr-2013-Keypoints from Symmetries by Wave Propagation
Author: Samuele Salti, Alessandro Lanza, Luigi Di_Stefano
Abstract: The paper conjectures and demonstrates that repeatable keypoints based on salient symmetries at different scales can be detected by a novel analysis grounded on the wave equation rather than the heat equation underlying traditional Gaussian scale–space theory. While the image structures found by most state-of-the-art detectors, such as blobs and corners, occur typically on planar highly textured surfaces, salient symmetries are widespread in diverse kinds of images, including those related to untextured objects, which are hardly dealt with by current feature-based recognition pipelines. We provide experimental results on standard datasets and also contribute with a new dataset focused on untextured objects. Based on the positive experimental results, we hope to foster further research on the promising topic ofscale invariant analysis through the wave equation.
2 0.72482061 298 cvpr-2013-Multi-scale Curve Detection on Surfaces
Author: Michael Kolomenkin, Ilan Shimshoni, Ayellet Tal
Abstract: This paper extends to surfaces the multi-scale approach of edge detection on images. The common practice for detecting curves on surfaces requires the user to first select the scale of the features, apply an appropriate smoothing, and detect the edges on the smoothed surface. This approach suffers from two drawbacks. First, it relies on a hidden assumption that all the features on the surface are of the same scale. Second, manual user intervention is required. In this paper, we propose a general framework for automatically detecting the optimal scale for each point on the surface. We smooth the surface at each point according to this optimal scale and run the curve detection algorithm on the resulting surface. Our multi-scale algorithm solves the two disadvantages of the single-scale approach mentioned above. We demonstrate how to realize our approach on two commonly-used special cases: ridges & valleys and relief edges. In each case, the optimal scale is found in accordance with the mathematical definition of the curve.
3 0.69233155 234 cvpr-2013-Joint Spectral Correspondence for Disparate Image Matching
Author: Mayank Bansal, Kostas Daniilidis
Abstract: We address the problem of matching images with disparate appearance arising from factors like dramatic illumination (day vs. night), age (historic vs. new) and rendering style differences. The lack of local intensity or gradient patterns in these images makes the application of pixellevel descriptors like SIFT infeasible. We propose a novel formulation for detecting and matching persistent features between such images by analyzing the eigen-spectrum of the joint image graph constructed from all the pixels in the two images. We show experimental results of our approach on a public dataset of challenging image pairs and demonstrate significant performance improvements over state-of-the-art.
4 0.64105725 437 cvpr-2013-Towards Fast and Accurate Segmentation
Author: Camillo Jose Taylor
Abstract: In this paper we explore approaches to accelerating segmentation and edge detection algorithms based on the gPb framework. The paper characterizes the performance of a simple but effective edge detection scheme which can be computed rapidly and offers performance that is competitive with the pB detector. The paper also describes an approach for computing a reduced order normalized cut that captures the essential features of the original problem but can be computed in less than half a second on a standard computing platform.
5 0.63276029 327 cvpr-2013-Pattern-Driven Colorization of 3D Surfaces
Author: George Leifman, Ayellet Tal
Abstract: Colorization refers to the process of adding color to black & white images or videos. This paper extends the term to handle surfaces in three dimensions. This is important for applications in which the colors of an object need to be restored and no relevant image exists for texturing it. We focus on surfaces with patterns and propose a novel algorithm for adding colors to these surfaces. The user needs only to scribble a few color strokes on one instance of each pattern, and the system proceeds to automatically colorize the whole surface. For this scheme to work, we address not only the problem of colorization, but also the problem of pattern detection on surfaces.
6 0.61827379 52 cvpr-2013-Axially Symmetric 3D Pots Configuration System Using Axis of Symmetry and Break Curve
7 0.61522031 429 cvpr-2013-The Generalized Laplacian Distance and Its Applications for Visual Matching
8 0.61106241 184 cvpr-2013-Gauging Association Patterns of Chromosome Territories via Chromatic Median
9 0.60521263 106 cvpr-2013-Deformable Graph Matching
10 0.59557396 222 cvpr-2013-Incorporating User Interaction and Topological Constraints within Contour Completion via Discrete Calculus
11 0.59407544 11 cvpr-2013-A Genetic Algorithm-Based Solver for Very Large Jigsaw Puzzles
12 0.58230746 72 cvpr-2013-Boundary Detection Benchmarking: Beyond F-Measures
13 0.57811534 91 cvpr-2013-Consensus of k-NNs for Robust Neighborhood Selection on Graph-Based Manifolds
14 0.57770288 350 cvpr-2013-Reconstructing Loopy Curvilinear Structures Using Integer Programming
15 0.56620008 468 cvpr-2013-Winding Number for Region-Boundary Consistent Salient Contour Extraction
16 0.5636698 373 cvpr-2013-SWIGS: A Swift Guided Sampling Method
17 0.558568 126 cvpr-2013-Diffusion Processes for Retrieval Revisited
18 0.55561858 162 cvpr-2013-FasT-Match: Fast Affine Template Matching
19 0.54848361 141 cvpr-2013-Efficient Computation of Shortest Path-Concavity for 3D Meshes
20 0.54760969 140 cvpr-2013-Efficient Color Boundary Detection with Color-Opponent Mechanisms
topicId topicWeight
[(10, 0.087), (16, 0.021), (26, 0.036), (28, 0.03), (33, 0.216), (39, 0.301), (59, 0.013), (67, 0.072), (69, 0.056), (87, 0.079), (93, 0.01)]
simIndex simValue paperId paperTitle
1 0.81979609 381 cvpr-2013-Scene Parsing by Integrating Function, Geometry and Appearance Models
Author: Yibiao Zhao, Song-Chun Zhu
Abstract: Indoor functional objects exhibit large view and appearance variations, thus are difficult to be recognized by the traditional appearance-based classification paradigm. In this paper, we present an algorithm to parse indoor images based on two observations: i) The functionality is the most essentialproperty to define an indoor object, e.g. “a chair to sit on ”; ii) The geometry (3D shape) ofan object is designed to serve its function. We formulate the nature of the object function into a stochastic grammar model. This model characterizes a joint distribution over the function-geometryappearance (FGA) hierarchy. The hierarchical structure includes a scene category, , functional groups, , functional objects, functional parts and 3D geometric shapes. We use a simulated annealing MCMC algorithm to find the maximum a posteriori (MAP) solution, i.e. a parse tree. We design four data-driven steps to accelerate the search in the FGA space: i) group the line segments into 3D primitive shapes, ii) assign functional labels to these 3D primitive shapes, iii) fill in missing objects/parts according to the functional labels, and iv) synthesize 2D segmentation maps and verify the current parse tree by the Metropolis-Hastings acceptance probability. The experimental results on several challenging indoor datasets demonstrate theproposed approach not only significantly widens the scope ofindoor sceneparsing algorithm from the segmentation and the 3D recovery to the functional object recognition, but also yields improved overall performance.
same-paper 2 0.79499167 240 cvpr-2013-Keypoints from Symmetries by Wave Propagation
Author: Samuele Salti, Alessandro Lanza, Luigi Di_Stefano
Abstract: The paper conjectures and demonstrates that repeatable keypoints based on salient symmetries at different scales can be detected by a novel analysis grounded on the wave equation rather than the heat equation underlying traditional Gaussian scale–space theory. While the image structures found by most state-of-the-art detectors, such as blobs and corners, occur typically on planar highly textured surfaces, salient symmetries are widespread in diverse kinds of images, including those related to untextured objects, which are hardly dealt with by current feature-based recognition pipelines. We provide experimental results on standard datasets and also contribute with a new dataset focused on untextured objects. Based on the positive experimental results, we hope to foster further research on the promising topic ofscale invariant analysis through the wave equation.
3 0.78759277 136 cvpr-2013-Discriminatively Trained And-Or Tree Models for Object Detection
Author: Xi Song, Tianfu Wu, Yunde Jia, Song-Chun Zhu
Abstract: This paper presents a method of learning reconfigurable And-Or Tree (AOT) models discriminatively from weakly annotated data for object detection. To explore the appearance and geometry space of latent structures effectively, we first quantize the image lattice using an overcomplete set of shape primitives, and then organize them into a directed acyclic And-Or Graph (AOG) by exploiting their compositional relations. We allow overlaps between child nodes when combining them into a parent node, which is equivalent to introducing an appearance Or-node implicitly for the overlapped portion. The learning of an AOT model consists of three components: (i) Unsupervised sub-category learning (i.e., branches of an object Or-node) with the latent structures in AOG being integrated out. (ii) Weaklysupervised part configuration learning (i.e., seeking the globally optimal parse trees in AOG for each sub-category). To search the globally optimal parse tree in AOG efficiently, we propose a dynamic programming (DP) algorithm. (iii) Joint appearance and structural parameters training under latent structural SVM framework. In experiments, our method is tested on PASCAL VOC 2007 and 2010 detection , benchmarks of 20 object classes and outperforms comparable state-of-the-art methods.
4 0.73364371 30 cvpr-2013-Accurate Localization of 3D Objects from RGB-D Data Using Segmentation Hypotheses
Author: Byung-soo Kim, Shili Xu, Silvio Savarese
Abstract: In this paper we focus on the problem of detecting objects in 3D from RGB-D images. We propose a novel framework that explores the compatibility between segmentation hypotheses of the object in the image and the corresponding 3D map. Our framework allows to discover the optimal location of the object using a generalization of the structural latent SVM formulation in 3D as well as the definition of a new loss function defined over the 3D space in training. We evaluate our method using two existing RGB-D datasets. Extensive quantitative and qualitative experimental results show that our proposed approach outperforms state-of-theart as methods well as a number of baseline approaches for both 3D and 2D object recognition tasks.
5 0.70873773 359 cvpr-2013-Robust Discriminative Response Map Fitting with Constrained Local Models
Author: Akshay Asthana, Stefanos Zafeiriou, Shiyang Cheng, Maja Pantic
Abstract: We present a novel discriminative regression based approach for the Constrained Local Models (CLMs) framework, referred to as the Discriminative Response Map Fitting (DRMF) method, which shows impressive performance in the generic face fitting scenario. The motivation behind this approach is that, unlike the holistic texture based features used in the discriminative AAM approaches, the response map can be represented by a small set of parameters and these parameters can be very efficiently used for reconstructing unseen response maps. Furthermore, we show that by adopting very simple off-the-shelf regression techniques, it is possible to learn robust functions from response maps to the shape parameters updates. The experiments, conducted on Multi-PIE, XM2VTS and LFPW database, show that the proposed DRMF method outperforms stateof-the-art algorithms for the task of generic face fitting. Moreover, the DRMF method is computationally very efficient and is real-time capable. The current MATLAB implementation takes 1second per image. To facilitate future comparisons, we release the MATLAB code1 and the pretrained models for research purposes.
7 0.69575942 397 cvpr-2013-Simultaneous Super-Resolution of Depth and Images Using a Single Camera
8 0.67645901 225 cvpr-2013-Integrating Grammar and Segmentation for Human Pose Estimation
9 0.67154169 446 cvpr-2013-Understanding Indoor Scenes Using 3D Geometric Phrases
10 0.66557157 61 cvpr-2013-Beyond Point Clouds: Scene Understanding by Reasoning Geometry and Physics
11 0.65873253 292 cvpr-2013-Multi-agent Event Detection: Localization and Role Assignment
12 0.65357476 461 cvpr-2013-Weakly Supervised Learning for Attribute Localization in Outdoor Scenes
13 0.64868611 248 cvpr-2013-Learning Collections of Part Models for Object Recognition
14 0.64767897 4 cvpr-2013-3D Visual Proxemics: Recognizing Human Interactions in 3D from a Single Image
15 0.64661723 365 cvpr-2013-Robust Real-Time Tracking of Multiple Objects by Volumetric Mass Densities
16 0.64604312 220 cvpr-2013-In Defense of Sparsity Based Face Recognition
17 0.64370078 119 cvpr-2013-Detecting and Aligning Faces by Image Retrieval
18 0.64190573 94 cvpr-2013-Context-Aware Modeling and Recognition of Activities in Video
19 0.64180404 242 cvpr-2013-Label Propagation from ImageNet to 3D Point Clouds
20 0.64127004 322 cvpr-2013-PISA: Pixelwise Image Saliency by Aggregating Complementary Appearance Contrast Measures with Spatial Priors