Co-segmentation of Non-homogeneous Image Sets
|
|
- Sybil Taylor
- 5 years ago
- Views:
Transcription
1 1 Co-segmentation of Non-homogeneous Image Sets Avik Hati Subhasis Chaudhuri Rajbabu Velmurugan Abstract In this paper, we formulate image co-segmentation as a classification problem in an unsupervised framework with the classes being the common foreground and the remaining regions in the image set. We first find a set of superpixels across all images with high feature similarity such that the constituent superpixels in individual images are spatially compact and label them as seed for the common foreground. Those superpixels with high background probability are labeled as respective seeds for multiple background classes. Seed computation here is unsupervised and automated unlike some semisupervised methods. Then, we compute discriminative features that separate the initially labeled classes using linear discriminant analysis. We use these features to perform spatially constrained label propagation and obtain labels for the unlabeled regions, and iterate this process till the seed regions grow to the common object. Experimental results demonstrate excellent robustness properties even while processing non-homogeneous image sets where the common object is present only in majority of the images. Co-segmentation, outliers, superpixel Index Terms I. INTRODUCTION Image co-segmentation [1], [2] is the process of finding the common object(s) with similar features from a set of two or more images by simultaneously segmenting all the images. See Fig. 1 for illustration. The images to be co-segmented are typically captured under different camera illumination and compositional context (e.g., crowd-sourced images). These variations make co-segmentation extremely difficult. Further in a non-homogeneous dataset, there may be some totally irrelevant images (outliers) that do not contain the common object. This makes co-segmentation even more challenging. Rother et al. [1] first introduced and solved the co-segmentation problem for an image pair and it was followed by the methods in [3], [4], [5] that proposed different optimization techniques for more efficient solutions. Subsequently, the focus A. Hati is the corresponding author ( avik.hati17@gmail.com). A. Hati, S. Chaudhuri and R. Velmurugan are with the Department of Electrical Engineering, Indian Institute of Technology Bombay, India.
2 2 Fig. 1: Co-segmentation result obtained using the proposed method. Row 1 shows a set of six inputs images to be co-segmented. Row 2 shows the co-segmentation output. shifted to co-segmentation of more than two images because of its many practical applications. Some of these works include the methods [6], [7], [8], [9], [10], [11], [12] that jointly segment the images into multiple classes to find foreground objects of different classes. But the methods in [6], [7], [9] require the common object class to be identified manually. Lee et al. [13] and Collins et al. [14] employed random walk for co-segmentation. The method in [15] co-segments different sized common foreground objects by constraining their histograms to have low entropy and to be linearly dependent. Tao et al. [16] use shape consistency among common foreground regions which may be invalid for natural images due to changes in pose and viewpoint. Supervised methods in [17], [18], [19] co-segment a set of similar images using scribbles drawn by users. The methods in [20], [21], [22], [23] first compute image saliency and use it as a key component in their co-segmentation methods. Meng et al. [24] and Chen et al. [21] split the input image set into simple and complex subsets. Subsequently they use results obtained from the simple subset for co-segmentation of the complex one. However, the common object may not always be salient in all constituent images. Learning based co-segmentation algorithms have been proposed in [25], [26], [27]. In order to obtain high-level features, semi-supervised methods in [28], [29], [30] compute region proposals from images using pre-trained networks, whereas Quan et al. [31] use CNN features. As mentioned earlier, the set of crowd-sourced images to be co-segmented may include outlier images that do not at all share the common object present in majority of the images in the set. Most existing methods [6], [7], [9], [11], [12], [13] do not consider this scenario ([32] and [20] do consider this). But the method of [20] being a saliency based approach cannot detect all non-salient co-segments. Wang et al. [32] first learn an object segmentation model from a training image set of same category. Consequently during co-segmentation, the outlier, if any, is rejected due to its non-conformability to the learned model. In this paper, we solve robust co-segmentation of an image set containing outlier images in a fully automated unsupervised framework, yet yielding excellent accuracy.
3 3 Input images Initial BG label & CFG label Compact cluster Superpixel clustering Discriminative features (LDA) Label propagation Common object in each image Fig. 2: Block diagram of the proposed co-segmentation algorithm (BG is background, CFG is common foreground). A. Problem definition In a set of images to be co-segmented, typically (i) the common object regions in different images are concentrated in the feature space since they are similar in feature and (ii) the background varies across images. But, variations in ambient imaging conditions across different images make feature selection difficult which in turn makes it very difficult to extract the common object(s) accurately. Hence, we propose to compute discriminative features (which better distinguish between background and the common foreground) using linear discriminant analysis (LDA) and formulate co-segmentation as a foregroundbackground classification problem. We first find a set of seed superpixels belonging to the common foreground class (C F ) and K B number of background classes (C Bk ) using an unsupervised method. Then we find labels for the remaining superpixels using a feature iterated label propagation method. The key contributions in this paper are The proposed method can handle outlier images. We show that discriminative feature computation through LDA in an unsupervised manner helps to reduce the limitations of low-level and mid-level features. We propose a label propagation technique with spatial constraints, that achieves robust co-segmentation. II. CO-SEGMENTATION ALGORITHM Let I 1, I 2,..., I N be the set of images to be co-segmented and the common object is present in M(< N) of them (M is unknown). First, we segment them into superpixels {s} using the simple linear iterative clustering (SLIC) method [33]. Then we compute low-level and mid-level features (Lab color, SIFT and bag-of-words feature) and define an object to be a set of contiguous superpixels with similar features. The proposed algorithm uses seed superpixel selection, discriminative feature computation and label propagation (LP) as shown in the block diagram in Fig. 2 and illustrated using Fig. 3, Fig. 4, Fig. 5.
4 4 Fig. 3: Illustration of superpixel clustering of six images including one outlier image. All superpixels from all images are clustered into R = 8 clusters based on their features. Row 1: Input image set to be co-segmented with I 1 being the outlier image. Rows 2-9: Clusters 1-8 and the sub-images constituted by the superpixels in every cluster. Cluster 5 (shown in Row 6) has the highest compactness value and superpixels from this cluster are chosen as seeds for the common foreground C F. A. Selection of superpixel seeds Typically, the set of superpixels constituting the common object in multiple images have similar features and they are closer to each other in the feature space compared to the background superpixels. Under this assumption, we first cluster all superpixels from all images based on their features. Let S ij be the superpixel subset of I i belonging to the j th cluster, assuming R clusters. Figure 3 shows an example considering R = 8 clusters. Since the common object superpixels in every image I i are spatially compact, we find the seeds for the common foreground C F as the set of superpixels belonging to the most spatially compact cluster. We define compactness of cluster-j as Γ j = ( N i=1 σ 2 ij A ij ) 1, where σ 2 ij is the spatial variance of centroids of superpixels in S ij and area A ij is the total number of pixels constituting them. For an image with a segmentable object, the superpixels should be spatially close, indicated by high Γ. Hence, we define C F = {s N i=1 S ip} where p = arg max j Γ j is the most compact cluster. The
5 5 Fig. 4: Illustration of seed labeling. Row 1: Input image set to be co-segmented. Row 2: Sub-images containing regions constituted by the common foreground (C F ) seed superpixels (in every image) that are obtained from the most compact cluster after superpixel feature clustering. Rows 3-6: Sub-images containing regions in every image constituted by the background (C Bk ) seed superpixels, ensuring P(s) > 0.99, in every image. sub-images constituted by C F is shown in Row 6 of Fig. 3 (and also in Row 2 of Fig. 4). We do not use information of remaining clusters further. To obtain seeds for the background, we use the background probability measure P( ), obtained from geodesic distance of every superpixel from image boundary [34]. We find the set of background seed superpixels with high confidence as S B = {s N i=1 I i : P(s) > 0.99}. As the superpixels in S B belong to different images, they usually have highly varying features. Grouping feature points with large variance together may lead to poor convergence during the LP stage. So, we cluster them into K B clusters (different from the clusters used of Γ computation) and obtain C B1, C B2,.... Hence, the total number of classes for labeling is K = 1 + K B. Rows 3-6 in Fig. 4 show background seed regions in four clusters. B. Feature iterated label propagation The set of seed superpixels are used to label the remaining superpixels. To achieve a more accurate labeling and robust co-segmentation, we first project the computed low-level and mid-level features (x R d ) onto a space that better discriminates foreground from background and obtain discriminative features. We use LDA to find the optimal discriminants and use them for feature projection [35] as
6 6 described next. For simplicity, let C 1 = C F, C 2 = C B1, C 3 = C B2,... Let X R d nt contains feature vectors of all (say n T ) labeled superpixels. Let (n k, m k ) be the cardinality of C k and the mean of feature vectors in C k, n T = K k=1 n k, and m be the mean of all feature vectors in all classes. We define the inter-class (or between-class) scatter matrix Q b R d d and the intra-class (or within-class) scatter matrix Q w R d d as in [35] and Q b = K k=1 n k n T (m k m)(m k m) T (1) Q w = K k=1 n k n T Q k, (2) where Q k is the covariance matrix of feature vectors in C k. We compute an appropriate projection matrix W = [w 1 w 2... w dr ], with w i R d and d r < d, by maximizing the following cost function (see [35] for details). cost(w) = tr{(v w ) 1 V b } such that w T i w i = 1, (3) where V w = W T Q w W and V b = W T Q b W are the intra-class and inter-class scatter matrices in the projected domain and tr{ } denotes trace. In Eqn (3), a large value of tr{v b } ensures good separation among classes and a small value of tr{v w } ensures less overlap among classes in the projected domain. The solution W, that contains the discriminants w i, is determined by the eigenvectors corresponding to the d r largest eigenvalues of Q 1 w Q b [35]. Here, max(d r ) = K 1 as Q 1 w Q b has rank at most equal to K 1. We project every feature vector x (labeled as well as unlabeled) as z = W T x. In Sec. II-A, we obtained labels of seed superpixels and our aim is to grow them to find the common object by assigning labels to the remaining superpixels. We perform region growing in two stages. First, we perform LP considering all superpixels (say n A ) from all images simultaneously using the discriminative features (z). Then we prune the updated superpixel labels in every image independently using spatial constraints as described next. We assign every seed superpixel s i belonging class C j, a label L = j and define a binary seed label matrix Y R na K Y(i, j) = { 1 if superpixel si has label L = j (4a) 0 otherwise. (4b) Here, Y has class information of only the seed superpixels. Our aim is to obtain an optimal label matrix L R na K with class information of all n A superpixels from all images.
7 7 First, we compute the feature similarity matrix S R na na where S ij is the negative exponential of the normalized distance between z i and z j and D is a diagonal matrix with D ii = j S ij. To obtain the optimal L, we minimize the following cost function n A i,j=1 1 S ij L i 1 L j 2 + α Dii Djj n A i=1 L i Y i 2 (5) where α > 0 is a regularization parameter and L i is the i th row of L. The first term (smoothness term) updates L using the similarity matrix S. Thus, labels are assigned to unlabeled superpixels through label propagation from the labeled superpixels. The second term (data fitting term) minimizes the difference between L and Y. It has been shown in [36], that minimization with respect to L yields where β 1 = L = β 1 (I β 2 D 1/2 SD 1/2 ) 1 Y (6) α 1+α, β 2 = 1 1+α. We obtain the label of s i as L = arg max j L (i, j), under constraints J1, J2. (7) Every row and column of Y correspond to a superpixel and a class, respectively. If the number of superpixels in one class C j is significantly large compared to the remaining classes (C k, k j), the columns of Y corresponding to C k s will be sparse. In such scenario, the solution to Eqn (7) will be biased towards C j. Hence, we normalize every column of Y by its L 0 -norm. Next, we update this solution using two constraints. J1: L (i, j) is a measure of similarity of superpixel s i to the set of superpixels with label j. If L (i, j) is small for all j = 1, 2,..., K (i.e., max j L (i, j) < t l ), these similarity values may lead to wrong label assignment; so we do not label that superpixel s i. We set t l to be median(l ). J2: The label update formulation in Eqn (5) does not use any spatial information of superpixels. Hence, every newly labeled superpixel (based only on feature similarity) may not be a neighbor of the seed regions in that sub-image belonging to a certain class and that sub-image may contain many discontiguous regions. But typically objects and background regions, e.g., sky, field and water body are contiguous regions. Hence we add a spatial constraint to Eqn (5) so that an unlabeled superpixel s i will be considered for assignment of label L = j using Eqn (7) only if it belongs to the first order spatial neighborhood of an already labeled region (with L = j) in that sub-image. Due to the above two constraints, only a limited number of superpixels in the spatial neighborhood of already labeled superpixels are assigned labels. After this label updating, we consider all labeled superpixels to again compute discriminative features from original feature vectors and perform label propagation again. We iterate these two stages alternately until convergence as shown in the block diagram in Fig. 2. The iteration converges if
8 8 Fig. 5: Co-segmentation output. Row 1: Input image set to be co-segmented. Row 2: Co-segmented objects constituted by the common foreground superpixels (in every image) that are obtained after convergence of feature iterated label propagation. Rows 3-6: the updated background clusters after convergence. either there is no more unlabeled superpixel left or labels no longer get updated. Row 2 in Fig. 5 shows the final co-segmentation result after convergence. As α in Eqn (5) is non-zero, initial labels of the labeled superpixels also get updated. We set α = 0.2. The proposed algorithm is given as a pseudocode in Algorithm 1. III. RESULTS Choice of features and datasets: We compute dense SIFT and CSIFT features from all images and encode them using locality-constrained linear coding [37], with the codebook size being 100, to obtain mid-level features. We have used L a b mean color feature (length 3) as low-level features. Hence, the feature dimension D = = 203. Unlike other methods [7], [13], we process all images in their original sizes and process all images simultaneously. We use the commonly used icoseg dataset [17] that contains images of 38 homogeneous classes and create a much larger dataset by embedding each class with several outlier images randomly chosen from other classes. Each set may have upto 30% of the data as outlier images. This dataset has an overwhelmingly large 570 sets containing a total of 9,563 non-unique images (given in the supplementary material). Similarly, we have created another non-homogeneous dataset from the MIT object discovery
9 9 TABLE I: Comparison of Jaccard similarity (J) of the proposed method (PM) with [6], [7], [9], [11], [13], [20] on the non-homogeneous datasets created using the icoseg dataset and the MIT dataset. J \ Methods PM [13] [11] [9] [20] [6] [7] icoseg MIT dataset internet dataset [20] for experiments. We have computed results for other methods using the codes provided by the authors. A. Quantitative results We have used Jaccard similarity (J) [20] as the metric to quantitatively evaluate the performance of the proposed method (PM) and compare with state-of-the-art methods in [6], [7], [9], [11], [13], [20], which fall into the same class of unsupervised co-segmentation. Jaccard similarity is defined as the intersection over union between the ground-truth and the binary mask of the co-segmentation output. We compare the values of J obtained from 570 sets of images, with each set containing 5-50 images, using PM and the existing methods in Table I. For the methods in [6], [7], [9], output segmentation class has to be manually chosen for each dataset. The poor performance of the method in [20] is due to the use of saliency as a cue as discussed in Sec. I. We found the number of initial clusters R [6, 8] and background classes K B [3, 4] yield similar results. The quantitative analysis clearly shows the superiority of PM. These results are obtained without performing any post-processing (e.g., GrabCut). B. Qualitative comparison In Fig. 6, we show visual comparison of the co-segmentation outputs obtained using PM with that of [20], [9], [6], [13], [11] on the pyramid images from the icoseg dataset that also includes one outlier image (Image 1) from goose subset. The proposed method correctly co-segments only pyramids (Row 7), whereas the methods in [9], [6], [13], [11] (Rows 3-6) incorrectly detect some background regions and horse in I 4, I 5, I 7, and the method in [20] (Row 2) incorrectly detects the sky as the common object. Moreover, they cannot handle the presence of outlier image and wrongly co-segment part of the goose (regions of significant size in case of [13]) from them unlike PM. More results are shown in Fig. 8, Fig. 9, Fig. 10, Fig. 11. In Fig. 7, we show that the proposed method can detect common objects of multiple classes by selecting foreground seeds from the second most compact cluster defined in Sec. II-A, after having detected the common object belonging to the previous class.
10 10 Fig. 6: Visual comparison of co-segmentation results on a set of 7 images (Row 1) from the icoseg dataset with I1 being the outlier. Results obtained using methods in [20], [9], [6], [13], [11] and PM are shown in Rows 2-7, respectively. Row 8: ground-truth. IV. C ONCLUSIONS We propose a co-segmentation algorithm that finds the common objects in an image set using a spatially constrained label propagation (LP) method that grows initially computed seed regions by assigning labels to the remaining regions. We do not use high level information like region proposals, saliency or CNN features that are used in semi-supervised methods. Instead we used only low and mid-level features to find seeds, which are used to obtain discriminative features for better feature representation in a completely unsupervised manner. LP in conjunction with LDA results in a more accurate labeling, thus yielding robust segmentation. The spatial constraint at the LP stage ensures spatially compact co-segmented objects.
11 11 Fig. 7: Illustration of multiple class co-segmentation. Row 1: Input image set. Rows 2, 3: Common objects in class 1 (helicopter) and class 2 (balloon), respectively. Fig. 8: Row 1: Input image set to be co-segmented with I6 being the outlier image. Rows 2-6: Cosegmentation output of methods in [20], [9], [6], [13], [11], respectively. Row 7: Co-segmentation output of the proposed method. Row 8: Ground-truth. R EFERENCES [1] C. Rother, T. Minka, A. Blake, and V. Kolmogorov, Cosegmentation of image pairs by histogram matching - incorporating a global constraint into MRFs, in Proc. CVPR, June 2006, vol. 1, pp
12 12 Fig. 9: Row 1: Input image set to be co-segmented with I7 being the outlier image. Rows 2-6: Cosegmentation output of methods in [20], [9], [6], [13], [11], respectively. Row 7: Co-segmentation output of the proposed method. Row 8: Ground-truth. [2] A. Hati, S. Chaudhuri, and R. Velmurugan, Image co-segmentation using maximum common subgraph matching and region co-growing, in Proc. ECCV, 2016, pp [3] L. Mukherjee, V. Singh, and C. Dyer, Half-integrality based algorithms for cosegmentation of images, in Proc. CVPR, June 2009, pp [4] D. Hochbaum and V. Singh, An efficient algorithm for co-segmentation, in Proc. ICCV, Sept 2009, pp [5] S. Vicente, V. Kolmogorov, and C. Rother, Cosegmentation revisited: Models and optimization, in Proc. ECCV, 2010, pp [6] A. Joulin, F. Bach, and J. Ponce, Discriminative clustering for image co-segmentation, in Proc. CVPR, June 2010, pp [7] G. Kim, E. Xing, L. Fei-Fei, and T. Kanade, Distributed cosegmentation via submodular optimization on anisotropic diffusion, in Proc. ICCV, Nov 2011, pp
13 13 Fig. 10: Row 1: Input image set to be co-segmented with I1 being the outlier image. Rows 2-6: Cosegmentation output of methods in [20], [9], [6], [13], [11], respectively. Row 7: Co-segmentation output of the proposed method. Row 8: Ground-truth. [8] G. Kim and E. Xing, On multiple foreground cosegmentation, in Proc. CVPR, June 2012, pp [9] A. Joulin, F. Bach, and J. Ponce, Multi-class cosegmentation, in Proc. CVPR, June 2012, pp [10] F. Wang, Q. Huang, M. Ovsjanikov, and L. J. Guibas, Unsupervised multi-class joint image segmentation, in Proc. CVPR, 2014, pp [11] H.-S. Chang and Y.-C. F. Wang, Optimizing the decomposition for multiple foreground cosegmentation, Elsevier Comput. Vis. Image Understand., vol. 141, pp , [12] J. Ma, S. Li, H. Qin, and A. Hao, Unsupervised multi-class co-segmentation via joint-cut over L1 -manifold hyper-graph of discriminative image regions, IEEE Trans. Image Process., vol. 26, no. 3, pp , [13] C. Lee, W.-D. Jang, J.-Y. Sim, and C.-S. Kim, Multiple random walkers and their application to image cosegmentation, in Proc. CVPR, June 2015, pp [14] M. Collins, J. Xu, L. Grady, and V. Singh, Random walks based multi-image segmentation: Quasiconvexity results and
14 14 Fig. 11: Row 1: Input image set to be co-segmented with I1 being the outlier image. Rows 2-6: Cosegmentation output of methods in [20], [9], [6], [13], [11], respectively. Row 7: Co-segmentation output of the proposed method. Row 8: Ground-truth. GPU-based solutions, in Proc. CVPR, June 2012, pp [15] L. Mukherjee, V. Singh, and J. Peng, Scale invariant cosegmentation for image groups, in Proc. CVPR, June 2011, pp [16] W. Tao, K. Li, and K. Sun, Sacoseg: Object cosegmentation by shape conformability, IEEE Trans. Image Process., vol. 24, no. 3, pp , [17] D. Batra, A. Kowdle, D. Parikh, J. Luo, and T. Chen, icoseg: Interactive co-segmentation with intelligent scribble guidance, in Proc. CVPR, June 2010, pp [18] X. Dong, J. Shen, L. Shao, and M.-H. Yang, Interactive cosegmentation using global and local energy optimization, IEEE Trans. Image Process., vol. 24, no. 11, pp , [19] W. Wang and J. Shen, Higher-order image co-segmentation, IEEE Trans. Multimedia, vol. 18, no. 6, pp , 2016.
15 15 [20] M. Rubinstein, A. Joulin, J. Kopf, and C. Liu, Unsupervised joint object discovery and segmentation in internet images, in Proc. CVPR, June 2013, pp [21] M. Chen, S. Velasco-Forero, I. Tsang, and T.-J. Cham, Objects co-segmentation: Propagated from simpler images, in Proc. ICASSP, 2015, pp [22] L. Lattari, A. Montenegro, and C. Vasconcelos, Unsupervised cosegmentation based on global clustering and saliency, in Proc. ICIP, 2015, pp [23] K. R. Jerripothula, J. Cai, and J. Yuan, Image co-segmentation via saliency co-fusion, IEEE Trans. Multimedia, vol. 18, no. 9, pp , [24] F. Meng, J. Cai, and H. Li, Cosegmentation of multiple image groups, Elsevier Comput. Vis. Image Understand., vol. 146, pp , [25] J. Sun and J. Ponce, Learning dictionary of discriminative part detectors for image categorization and cosegmentation, Intl. Journal Comput. Vis., vol. 120, no. 2, pp , [26] J. C. Rubio, J. Serrat, A. López, and N. Paragios, Unsupervised co-segmentation through region matching, in Proc. CVPR, 2012, pp [27] Z. Yuan, T. Lu, and Y. Wu, Deep-dense conditional random fields for object co-segmentation, in Proc. IJCAI, 2017, pp [28] S. Vicente, C. Rother, and V. Kolmogorov, Object cosegmentation, in Proc. CVPR, June 2011, pp [29] K. Li, J. Zhang, and W. Tao, Unsupervised co-segmentation for indefinite number of common foreground objects, IEEE Trans. Image Process., vol. 25, no. 4, pp , [30] Y. Li, L. Liu, C. Shen, and A. van den Hengel, Image co-localization by mimicking a good detectors confidence score distribution, in Proc. ECCV, 2016, pp [31] R. Quan, J. Han, D. Zhang, and F. Nie, Object co-segmentation via graph optimized-flexible manifold ranking, in Proc. CVPR, 2016, pp [32] L. Wang, G. Hua, J. Xue, Z. Gao, and N. Zheng, Joint segmentation and recognition of categorized objects from noisy web image collection, IEEE Trans. Image Process., vol. 23, no. 9, pp , [33] R. Achanta, A. Shaji, K. Smith, A. Lucchi, P. Fua, and S. Süsstrunk, SLIC superpixels compared to state-of-the-art superpixel methods, IEEE Trans. PAMI, vol. 34, no. 11, pp , [34] W. Zhu, S. Liang, Y. Wei, and J. Sun, Saliency optimization from robust background detection, in Proc. CVPR, 2014, pp [35] C. M. Bishop, Pattern recognition and machine learning, Springer, [36] D. Zhou, O. Bousquet, T. N. Lal, J. Weston, and B. Schölkopf, Learning with local and global consistency, in Advances in NIPS, 2004, pp [37] J. Wang, J. Yang, K. Yu, F. Lv, T. Huang, and Y. Gong, Locality-constrained linear coding for image classification, in Proc. CVPR, 2010, pp
16 Algorithm 1 Proposed co-segmentation algorithm Input: Set of images I 1, I 2,..., I N Output: Set of superpixels in all images belonging to the common objects 1: for i = 1 to N do 2: Superpixel segmentation of every image I i 3: Compute background probability P(r) of every superpixel r I i 4: end for 5: Compute feature x for every suprepixel r using LLC from SIFT, CSIFT and L a b mean color 6: // Initial labeling 7: Cluster all superpixels {x N i=1 I i} from all images into R clusters based on their features 8: for j = 1 to R do 9: for i = 1 to N do 10: S ij superpixel subset of I i belonging to the j th cluster 11: A ij total number of pixels constituting the superpixels in S ij 12: σij 2 spatial variance of centroids of superpixels in S ij 13: end for 14: Compute compactness of every cluster Γ j 15: end for ( N i=1 ) σ 2 1 ij A ij 16: Find the most compact cluster p arg max j Γ j 17: Common foreground seed C F {r N i=1 S ip} 18: Cluster {r N i=1 I i : P(r) > 0.99} in K B clusters and find background seeds as clusters C B1, C B2, C B3,... 19: K K B : Initialize t 0, C (0) 1 C F, C (0) 2 C B1, C (0) 3 C B2, C (0) 4 C B3,... 21: For each k = 1, 2,..., K, assign label L (0) r k, r C k 22: // Label propagation 23: while no convergence do 24: Find discriminant matrix W from features in C (t) 1, C(t) 2, C(t) 3,... using LDA 25: Project every feature vector x (labeled as well as unlabeled) as z W T x 26: Compute similarity matrix S where S(i, j) similarity(z i, z j ) 27: Compute diagonal matrix D where D(i, i) j S(i, j) 28: for k = 1 to K do 29: for r i C (t) k do 30: Y(i, k) 1 C (t) k 31: end for 32: end for 33: L (I αd 1/2 SD 1/2 ) 1 Y 34: t l median(y ) 35: // Label update 36: for all r i do 37: if max k 38: L (t+1) r i L (i, k) > t l r i N s ( K arg max L (i, k) k 39: end if 40: end for 41: For each k = 1, 2,..., K, C (t+1) k k=1 C(t) k ) then {r : L (t+1) r = k} 42: t t : end while 44: C F = C 1 at convergence is the set of superpixels in all images constituting the common object 16
UNSUPERVISED CO-SEGMENTATION BASED ON A NEW GLOBAL GMM CONSTRAINT IN MRF. Hongkai Yu, Min Xian, and Xiaojun Qi
UNSUPERVISED CO-SEGMENTATION BASED ON A NEW GLOBAL GMM CONSTRAINT IN MRF Hongkai Yu, Min Xian, and Xiaojun Qi Department of Computer Science, Utah State University, Logan, UT 84322-4205 hongkai.yu@aggiemail.usu.edu,
More informationAUTOMATIC IMAGE CO-SEGMENTATION USING GEOMETRIC MEAN SALIENCY
AUTOMATIC IMAGE CO-SEGMENTATION USING GEOMETRIC MEAN SALIENCY Koteswar Rao Jerripothula Jianfei Cai Fanman Meng Junsong Yuan ROSE Lab, Interdisciplinary Graduate School, Nanyang Technological University,
More informationAn Efficient Image Co-segmentation Algorithm based on Active Contour and Image Saliency
An Efficient Image Co-segmentation Algorithm based on Active Contour and Image Saliency Zhizhi Zhang 1, Xiabi Liu 1, Nouman Qadeer Soomro 2 1 Beijing Laboratory of Intelligent Information Technology, School
More informationUNSUPERVISED COSEGMENTATION BASED ON SUPERPIXEL MATCHING AND FASTGRABCUT. Hongkai Yu and Xiaojun Qi
UNSUPERVISED COSEGMENTATION BASED ON SUPERPIXEL MATCHING AND FASTGRABCUT Hongkai Yu and Xiaojun Qi Department of Computer Science, Utah State University, Logan, UT 84322-4205 hongkai.yu@aggiemail.usu.edu
More informationSupplementary Materials for Salient Object Detection: A
Supplementary Materials for Salient Object Detection: A Discriminative Regional Feature Integration Approach Huaizu Jiang, Zejian Yuan, Ming-Ming Cheng, Yihong Gong Nanning Zheng, and Jingdong Wang Abstract
More informationDiscriminative Clustering for Image Co-Segmentation
Discriminative Clustering for Image Co-Segmentation Joulin, A.; Bach, F.; Ponce, J. (CVPR. 2010) Iretiayo Akinola Josh Tennefoss Outline Why Co-segmentation? Previous Work Problem Formulation Experimental
More informationImage Cosegmentation via Saliency-Guided Constrained Clustering with Cosine Similarity
Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17) Image Cosegmentation via Saliency-Guided Constrained Clustering with Cosine Similarity Zhiqiang Tao, 1 Hongfu Liu, 1
More informationarxiv: v1 [cs.cv] 17 Apr 2018
arxiv:1804.06423v1 [cs.cv] 17 Apr 2018 Deep Object Co-Segmentation Weihao Li, Omid Hosseini Jafari, Carsten Rother Visual Learning Lab Heidelberg University (HCI/IWR) http://vislearn.de Abstract. This
More informationMultiple cosegmentation
Armand Joulin, Francis Bach and Jean Ponce. INRIA -Ecole Normale Supérieure April 25, 2012 Segmentation Introduction Segmentation Supervised and weakly-supervised segmentation Cosegmentation Segmentation
More informationA Hierarchical Image Clustering Cosegmentation Framework
A Hierarchical Image Clustering Cosegmentation Framework Edward Kim, Hongsheng Li, Xiaolei Huang Department of Computer Science and Engineering Lehigh University, PA 18015 {edk208,h.li,xih206}@lehigh.edu
More informationSemi-supervised Data Representation via Affinity Graph Learning
1 Semi-supervised Data Representation via Affinity Graph Learning Weiya Ren 1 1 College of Information System and Management, National University of Defense Technology, Changsha, Hunan, P.R China, 410073
More informationAn efficient face recognition algorithm based on multi-kernel regularization learning
Acta Technica 61, No. 4A/2016, 75 84 c 2017 Institute of Thermomechanics CAS, v.v.i. An efficient face recognition algorithm based on multi-kernel regularization learning Bi Rongrong 1 Abstract. A novel
More informationUsing the Kolmogorov-Smirnov Test for Image Segmentation
Using the Kolmogorov-Smirnov Test for Image Segmentation Yong Jae Lee CS395T Computational Statistics Final Project Report May 6th, 2009 I. INTRODUCTION Image segmentation is a fundamental task in computer
More informationCosegmentation Revisited: Models and Optimization
Cosegmentation Revisited: Models and Optimization Sara Vicente 1, Vladimir Kolmogorov 1, and Carsten Rother 2 1 University College London 2 Microsoft Research Cambridge Abstract. The problem of cosegmentation
More informationSupervised texture detection in images
Supervised texture detection in images Branislav Mičušík and Allan Hanbury Pattern Recognition and Image Processing Group, Institute of Computer Aided Automation, Vienna University of Technology Favoritenstraße
More informationSALIENT OBJECT DETECTION FOR RGB-D IMAGE VIA SALIENCY EVOLUTION.
SALIENT OBJECT DETECTION FOR RGB-D IMAGE VIA SALIENCY EVOLUTION Jingfan Guo,2, Tongwei Ren,2,, Jia Bei,2 State Key Laboratory for Novel Software Technology, Nanjing University, China 2 Software Institute,
More informationObject Co-Segmentation Based on Shortest Path Algorithm and Saliency Model
IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 14, NO. 5, OCTOBER 2012 1429 Object Co-Segmentation Based on Shortest Path Algorithm and Saliency Model Fanman Meng, Hongliang Li, Senior Member, IEEE, Guanghui Liu,
More informationData-driven Saliency Region Detection Based on Undirected Graph Ranking
Data-driven Saliency Region Detection Based on Undirected Graph Ranking Wenjie Zhang ; Qingyu Xiong 2 ; Shunhan Chen 3 College of Automation, 2 the School of Software Engineering, 3 College of Information
More informationImage Co-Segmentation via Consistent Functional Maps
Image Co-Segmentation via Consistent Functional Maps Fan Wang Stanford University fanw@stanford.edu Qixing Huang Stanford University huangqx@stanford.edu Leonidas J. Guibas Stanford University guibas@cs.stanford.edu
More informationAn ICA based Approach for Complex Color Scene Text Binarization
An ICA based Approach for Complex Color Scene Text Binarization Siddharth Kherada IIIT-Hyderabad, India siddharth.kherada@research.iiit.ac.in Anoop M. Namboodiri IIIT-Hyderabad, India anoop@iiit.ac.in
More informationCo-Saliency Detection within a Single Image
Co-Saliency Detection within a Single Image Hongkai Yu2, Kang Zheng2, Jianwu Fang3, 4, Hao Guo2, Wei Feng, Song Wang, 2, 2 School of Computer Science and Technology, Tianjin University, Tianjin, China
More informationSCALP: Superpixels with Contour Adherence using Linear Path
SCALP: Superpixels with Contour Adherence using Linear Path Rémi Giraud 1,2 remi.giraud@labri.fr with Vinh-Thong Ta 1 and Nicolas Papadakis 2 1 LaBRI CNRS UMR 5800 - University of Bordeaux, FRANCE PICTURA
More informationRobust interactive image segmentation via graph-based manifold ranking
Computational Visual Media DOI 10.1007/s41095-015-0024-2 Vol. 1, No. 3, September 2015, 183 195 Research Article Robust interactive image segmentation via graph-based manifold ranking Hong Li 1 ( ), Wen
More informationUnsupervised Joint Object Discovery and Segmentation in Internet Images
Unsupervised Joint Object Discovery and Segmentation in Internet Images Michael Rubinstein1,3 1 Car Armand Joulin2,3 MIT CSAIL Image search 2 INRIA 3 Johannes Kopf3 Ce Liu3 Microsoft Research Object discovery
More informationACCORDING to the principle of human visual perception,
IEEE SIGNAL PROCESSING LETTERS, VOL. XX, NO. XX, XXXX 2016 1 Saliency Detection for Stereoscopic Images Based on Depth Confidence Analysis Multiple Cues Fusion Runmin Cong, Student Member, IEEE, Jianjun
More informationEnhanced and Efficient Image Retrieval via Saliency Feature and Visual Attention
Enhanced and Efficient Image Retrieval via Saliency Feature and Visual Attention Anand K. Hase, Baisa L. Gunjal Abstract In the real world applications such as landmark search, copy protection, fake image
More informationVideo Object Co-Segmentation by Regulated Maximum Weight Cliques
Video Object Co-Segmentation by Regulated Maximum Weight Cliques Dong Zhang 1, Omar Javed 2 and Mubarak Shah 1 1 Center for Research in Computer Vision, UCF, Orlando, FL 32826 2 SRI International, Princeton,
More informationCO-SALIENCY DETECTION VIA HIERARCHICAL CONSISTENCY MEASURE
CO-SALIENCY DETECTION VIA HIERARCHICAL CONSISTENCY MEASURE Yonghua Zhang 1,3, Liang Li 1,3, Runmin Cong 2, Xiaojie Guo 1,3, Hui Xu 1,3, Jiawan Zhang 1,3, 1 School of Computer Software, Tianjin University,
More informationJoint Inference in Image Databases via Dense Correspondence. Michael Rubinstein MIT CSAIL (while interning at Microsoft Research)
Joint Inference in Image Databases via Dense Correspondence Michael Rubinstein MIT CSAIL (while interning at Microsoft Research) My work Throughout the year (and my PhD thesis): Temporal Video Analysis
More informationFrequent Inner-Class Approach: A Semi-supervised Learning Technique for One-shot Learning
Frequent Inner-Class Approach: A Semi-supervised Learning Technique for One-shot Learning Izumi Suzuki, Koich Yamada, Muneyuki Unehara Nagaoka University of Technology, 1603-1, Kamitomioka Nagaoka, Niigata
More informationContent-based Image and Video Retrieval. Image Segmentation
Content-based Image and Video Retrieval Vorlesung, SS 2011 Image Segmentation 2.5.2011 / 9.5.2011 Image Segmentation One of the key problem in computer vision Identification of homogenous region in the
More informationIMAGE co-segmentation (e.g. [2], [3], [4], [5], [6]) is an
1 Image Co-segmentation via Multi-scale Local Shape Transfer Wei Teng, Yu Zhang, Xiaowu Chen, Jia Li, and Zhiqiang He arxiv:1805.05610v1 [cs.cv] 15 May 2018 Abstract Image co-segmentation is a challenging
More informationShape Co-analysis. Daniel Cohen-Or. Tel-Aviv University
Shape Co-analysis Daniel Cohen-Or Tel-Aviv University 1 High-level Shape analysis [Fu et al. 08] Upright orientation [Mehra et al. 08] Shape abstraction [Kalograkis et al. 10] Learning segmentation [Mitra
More informationGraph-Based Superpixel Labeling for Enhancement of Online Video Segmentation
Graph-Based Superpixel Labeling for Enhancement of Online Video Segmentation Alaa E. Abdel-Hakim Electrical Engineering Department Assiut University Assiut, Egypt alaa.aly@eng.au.edu.eg Mostafa Izz Cairo
More informationMarkov Random Fields and Segmentation with Graph Cuts
Markov Random Fields and Segmentation with Graph Cuts Computer Vision Jia-Bin Huang, Virginia Tech Many slides from D. Hoiem Administrative stuffs Final project Proposal due Oct 27 (Thursday) HW 4 is out
More informationUnsupervised Saliency Estimation based on Robust Hypotheses
Utah State University DigitalCommons@USU Computer Science Faculty and Staff Publications Computer Science 3-2016 Unsupervised Saliency Estimation based on Robust Hypotheses Fei Xu Utah State University,
More informationLearning based face hallucination techniques: A survey
Vol. 3 (2014-15) pp. 37-45. : A survey Premitha Premnath K Department of Computer Science & Engineering Vidya Academy of Science & Technology Thrissur - 680501, Kerala, India (email: premithakpnath@gmail.com)
More informationUnsupervised and Semi-Supervised Learning vial 1 -Norm Graph
Unsupervised and Semi-Supervised Learning vial -Norm Graph Feiping Nie, Hua Wang, Heng Huang, Chris Ding Department of Computer Science and Engineering University of Texas, Arlington, TX 769, USA {feipingnie,huawangcs}@gmail.com,
More informationSurvey: Recent Trends and Techniques in Image Co-Segmentation Challenges, Issues and Its Applications
ISSN (Online): 2409-4285 www.ijcsse.org Page: 99-114 Survey: Recent Trends and Techniques in Image Co-Segmentation Challenges, Issues and Its Applications Tri Daryanto 1, Sheeraz Arif 2 and Shu Yang 3
More informationAggregating Descriptors with Local Gaussian Metrics
Aggregating Descriptors with Local Gaussian Metrics Hideki Nakayama Grad. School of Information Science and Technology The University of Tokyo Tokyo, JAPAN nakayama@ci.i.u-tokyo.ac.jp Abstract Recently,
More informationSurvey On Segmentation And Recognition Of Categorized Objects
ISSN:2320-0790 Survey On Segmentation And Recognition Of Categorized Objects Nithina O, Prasanth Kumar P V PG scholar, Asst.Professor. Department of Computer Science Vimal Jyothi Engineering College, Kannur
More informationA TRAJECTORY CLUSTERING APPROACH TO CROWD FLOW SEGMENTATION IN VIDEOS. Rahul Sharma, Tanaya Guha
A TRAJECTORY CLUSTERING APPROACH TO CROWD FLOW SEGMENTATION IN VIDEOS Rahul Sharma, Tanaya Guha Electrical Engineering, Indian Institute of Technology Kanpur, India ABSTRACT This work proposes a trajectory
More informationIN COMPUTER vision area, image segmentation [1] [8]
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 22, NO. 12, DECEMBER 2013 4809 Feature Adaptive Co-Segmentation by Complexity Awareness Fanman Meng, Hongliang Li, Senior Member, IEEE, KingNgiNgan,Fellow, IEEE,
More informationDescriptorEnsemble: An Unsupervised Approach to Image Matching and Alignment with Multiple Descriptors
DescriptorEnsemble: An Unsupervised Approach to Image Matching and Alignment with Multiple Descriptors 林彥宇副研究員 Yen-Yu Lin, Associate Research Fellow 中央研究院資訊科技創新研究中心 Research Center for IT Innovation, Academia
More informationSupplementary Material for Ensemble Diffusion for Retrieval
Supplementary Material for Ensemble Diffusion for Retrieval Song Bai 1, Zhichao Zhou 1, Jingdong Wang, Xiang Bai 1, Longin Jan Latecki 3, Qi Tian 4 1 Huazhong University of Science and Technology, Microsoft
More informationObject Extraction Using Image Segmentation and Adaptive Constraint Propagation
Object Extraction Using Image Segmentation and Adaptive Constraint Propagation 1 Rajeshwary Patel, 2 Swarndeep Saket 1 Student, 2 Assistant Professor 1 2 Department of Computer Engineering, 1 2 L. J. Institutes
More informationClustering CS 550: Machine Learning
Clustering CS 550: Machine Learning This slide set mainly uses the slides given in the following links: http://www-users.cs.umn.edu/~kumar/dmbook/ch8.pdf http://www-users.cs.umn.edu/~kumar/dmbook/dmslides/chap8_basic_cluster_analysis.pdf
More informationUnsupervised discovery of category and object models. The task
Unsupervised discovery of category and object models Martial Hebert The task 1 Common ingredients 1. Generate candidate segments 2. Estimate similarity between candidate segments 3. Prune resulting (implicit)
More informationSemantic Context Forests for Learning- Based Knee Cartilage Segmentation in 3D MR Images
Semantic Context Forests for Learning- Based Knee Cartilage Segmentation in 3D MR Images MICCAI 2013: Workshop on Medical Computer Vision Authors: Quan Wang, Dijia Wu, Le Lu, Meizhu Liu, Kim L. Boyer,
More informationCo-attention CNNs for Unsupervised Object Co-segmentation
Co-attention CNNs for Unsupervised Object Co-segmentation Kuang-Jui Hsu 1,2, Yen-Yu Lin 1, Yung-Yu Chuang 1,2 1 Academia Sinica, Taiwan 2 National Taiwan University, Taiwan kjhsu@iis.sinica.edu.tw, yylin@citi.sinica.edu.tw,
More informationUnsupervised Outlier Detection and Semi-Supervised Learning
Unsupervised Outlier Detection and Semi-Supervised Learning Adam Vinueza Department of Computer Science University of Colorado Boulder, Colorado 832 vinueza@colorado.edu Gregory Z. Grudic Department of
More informationInteractive Image Segmentation with GrabCut
Interactive Image Segmentation with GrabCut Bryan Anenberg Stanford University anenberg@stanford.edu Michela Meister Stanford University mmeister@stanford.edu Abstract We implement GrabCut and experiment
More informationAnalysis: TextonBoost and Semantic Texton Forests. Daniel Munoz Februrary 9, 2009
Analysis: TextonBoost and Semantic Texton Forests Daniel Munoz 16-721 Februrary 9, 2009 Papers [shotton-eccv-06] J. Shotton, J. Winn, C. Rother, A. Criminisi, TextonBoost: Joint Appearance, Shape and Context
More informationTri-modal Human Body Segmentation
Tri-modal Human Body Segmentation Master of Science Thesis Cristina Palmero Cantariño Advisor: Sergio Escalera Guerrero February 6, 2014 Outline 1 Introduction 2 Tri-modal dataset 3 Proposed baseline 4
More informationCombining Selective Search Segmentation and Random Forest for Image Classification
Combining Selective Search Segmentation and Random Forest for Image Classification Gediminas Bertasius November 24, 2013 1 Problem Statement Random Forest algorithm have been successfully used in many
More informationObject detection using non-redundant local Binary Patterns
University of Wollongong Research Online Faculty of Informatics - Papers (Archive) Faculty of Engineering and Information Sciences 2010 Object detection using non-redundant local Binary Patterns Duc Thanh
More informationSegmentation in electron microscopy images
Segmentation in electron microscopy images Aurelien Lucchi, Kevin Smith, Yunpeng Li Bohumil Maco, Graham Knott, Pascal Fua. http://cvlab.epfl.ch/research/medical/neurons/ Outline Automated Approach to
More informationCAP 6412 Advanced Computer Vision
CAP 6412 Advanced Computer Vision http://www.cs.ucf.edu/~bgong/cap6412.html Boqing Gong April 21st, 2016 Today Administrivia Free parameters in an approach, model, or algorithm? Egocentric videos by Aisha
More informationSaliency Detection via Graph-Based Manifold Ranking
Saliency Detection via Graph-Based Manifold Ranking Chuan Yang, Lihe Zhang, Huchuan Lu, Xiang Ruan 2, and Ming-Hsuan Yang 3 Dalian University of Technology 2 OMRON Corporation 3 University of California
More informationRobust Face Recognition via Sparse Representation Authors: John Wright, Allen Y. Yang, Arvind Ganesh, S. Shankar Sastry, and Yi Ma
Robust Face Recognition via Sparse Representation Authors: John Wright, Allen Y. Yang, Arvind Ganesh, S. Shankar Sastry, and Yi Ma Presented by Hu Han Jan. 30 2014 For CSE 902 by Prof. Anil K. Jain: Selected
More informationFace Recognition Based on LDA and Improved Pairwise-Constrained Multiple Metric Learning Method
Journal of Information Hiding and Multimedia Signal Processing c 2016 ISSN 2073-4212 Ubiquitous International Volume 7, Number 5, September 2016 Face Recognition ased on LDA and Improved Pairwise-Constrained
More informationSupplementary Material: Unconstrained Salient Object Detection via Proposal Subset Optimization
Supplementary Material: Unconstrained Salient Object via Proposal Subset Optimization 1. Proof of the Submodularity According to Eqns. 10-12 in our paper, the objective function of the proposed optimization
More informationCluster Analysis. Mu-Chun Su. Department of Computer Science and Information Engineering National Central University 2003/3/11 1
Cluster Analysis Mu-Chun Su Department of Computer Science and Information Engineering National Central University 2003/3/11 1 Introduction Cluster analysis is the formal study of algorithms and methods
More informationDiscrete Optimization of Ray Potentials for Semantic 3D Reconstruction
Discrete Optimization of Ray Potentials for Semantic 3D Reconstruction Marc Pollefeys Joined work with Nikolay Savinov, Christian Haene, Lubor Ladicky 2 Comparison to Volumetric Fusion Higher-order ray
More informationCo-Saliency Detection Based on Hierarchical Segmentation
88 IEEE SIGNAL PROCESSING LETTERS, VOL. 21, NO. 1, JANUARY 2014 Co-Saliency Detection Based on Hierarchical Segmentation Zhi Liu, Member, IEEE, Wenbin Zou, Lina Li, Liquan Shen, and Olivier Le Meur Abstract
More information2 Proposed Methodology
3rd International Conference on Multimedia Technology(ICMT 2013) Object Detection in Image with Complex Background Dong Li, Yali Li, Fei He, Shengjin Wang 1 State Key Laboratory of Intelligent Technology
More informationImproving Image Segmentation Quality Via Graph Theory
International Symposium on Computers & Informatics (ISCI 05) Improving Image Segmentation Quality Via Graph Theory Xiangxiang Li, Songhao Zhu School of Automatic, Nanjing University of Post and Telecommunications,
More informationObject cosegmentation using deep Siamese network
Object cosegmentation using deep Siamese network Prerana Mukherjee, Brejesh Lall and Snehith Lattupally Dept of EE, IIT Delhi, India. Email: {eez138300, brejesh, eet152695 }@ee.iitd.ac.in arxiv:1803.02555v2
More informationDETECTION OF IMAGE PAIRS USING CO-SALIENCY MODEL
DETECTION OF IMAGE PAIRS USING CO-SALIENCY MODEL N S Sandhya Rani 1, Dr. S. Bhargavi 2 4th sem MTech, Signal Processing, S. J. C. Institute of Technology, Chickballapur, Karnataka, India 1 Professor, Dept
More informationSpectral Active Clustering of Remote Sensing Images
Spectral Active Clustering of Remote Sensing Images Zifeng Wang, Gui-Song Xia, Caiming Xiong, Liangpei Zhang To cite this version: Zifeng Wang, Gui-Song Xia, Caiming Xiong, Liangpei Zhang. Spectral Active
More informationGenerating Object Candidates from RGB-D Images and Point Clouds
Generating Object Candidates from RGB-D Images and Point Clouds Helge Wrede 11.05.2017 1 / 36 Outline Introduction Methods Overview The Data RGB-D Images Point Clouds Microsoft Kinect Generating Object
More informationFOREGROUND SEGMENTATION BASED ON MULTI-RESOLUTION AND MATTING
FOREGROUND SEGMENTATION BASED ON MULTI-RESOLUTION AND MATTING Xintong Yu 1,2, Xiaohan Liu 1,2, Yisong Chen 1 1 Graphics Laboratory, EECS Department, Peking University 2 Beijing University of Posts and
More informationDiscriminative Clustering for Image Co-segmentation
Discriminative Clustering for Image Co-segmentation Armand Joulin Francis Bach Jean Ponce INRIA Ecole Normale Supérieure, Paris January 2010 Introduction Introduction Task: dividing simultaneously q images
More informationSegmenting Objects in Weakly Labeled Videos
Segmenting Objects in Weakly Labeled Videos Mrigank Rochan, Shafin Rahman, Neil D.B. Bruce, Yang Wang Department of Computer Science University of Manitoba Winnipeg, Canada {mrochan, shafin12, bruce, ywang}@cs.umanitoba.ca
More informationString distance for automatic image classification
String distance for automatic image classification Nguyen Hong Thinh*, Le Vu Ha*, Barat Cecile** and Ducottet Christophe** *University of Engineering and Technology, Vietnam National University of HaNoi,
More informationOptimizing and Background Learning in a Single Process of Moving Object Detection
International Journal of Scientific Research in Computer Science, Engineering and Information Technology 2017 IJSRCSEIT Volume 2 Issue 2 ISSN : 2456-3307 Optimizing and Background Learning in a Single
More informationIllumination Estimation from Shadow Borders
Illumination Estimation from Shadow Borders Alexandros Panagopoulos, Tomás F. Yago Vicente, Dimitris Samaras Stony Brook University Stony Brook, NY, USA {apanagop, tyagovicente, samaras}@cs.stonybrook.edu
More informationSuperpixel Segmentation using Depth
Superpixel Segmentation using Depth Information Superpixel Segmentation using Depth Information David Stutz June 25th, 2014 David Stutz June 25th, 2014 01 Introduction - Table of Contents 1 Introduction
More informationData-driven Depth Inference from a Single Still Image
Data-driven Depth Inference from a Single Still Image Kyunghee Kim Computer Science Department Stanford University kyunghee.kim@stanford.edu Abstract Given an indoor image, how to recover its depth information
More informationCRFs for Image Classification
CRFs for Image Classification Devi Parikh and Dhruv Batra Carnegie Mellon University Pittsburgh, PA 15213 {dparikh,dbatra}@ece.cmu.edu Abstract We use Conditional Random Fields (CRFs) to classify regions
More informationImproving Latent Fingerprint Matching Performance by Orientation Field Estimation using Localized Dictionaries
Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 3, Issue. 11, November 2014,
More informationLOOSECUT: INTERACTIVE IMAGE SEGMENTATION WITH LOOSELY BOUNDED BOXES
Loose Input Box LOOSECUT: INTERACTIVE IMAGE SEGMENTATION WITH LOOSELY BOUNDED BOXES Hongkai Yu 1, Youjie Zhou 1, Hui Qian 2, Min Xian 3, and Song Wang 1 1 University of South Carolina, SC 2 Zhejiang University,
More informationMULTI-POSE FACE HALLUCINATION VIA NEIGHBOR EMBEDDING FOR FACIAL COMPONENTS. Yanghao Li, Jiaying Liu, Wenhan Yang, Zongming Guo
MULTI-POSE FACE HALLUCINATION VIA NEIGHBOR EMBEDDING FOR FACIAL COMPONENTS Yanghao Li, Jiaying Liu, Wenhan Yang, Zongg Guo Institute of Computer Science and Technology, Peking University, Beijing, P.R.China,
More informationarxiv: v1 [cs.lg] 20 Dec 2013
Unsupervised Feature Learning by Deep Sparse Coding Yunlong He Koray Kavukcuoglu Yun Wang Arthur Szlam Yanjun Qi arxiv:1312.5783v1 [cs.lg] 20 Dec 2013 Abstract In this paper, we propose a new unsupervised
More informationParallel Cosegmentation via Submodular Optimization on Anisotropic Diffusion
Parallel Cosegmentation via Submodular Optimization on Anisotropic Diffusion Dinesh Majeti, Aditya Prakash, S. Balasubramanian, PK Baruah Sri Sathya Sai Institute of Higher Learning, Prasanthi Nilayam,India.
More informationarxiv: v1 [cs.cv] 6 Jul 2016
arxiv:607.079v [cs.cv] 6 Jul 206 Deep CORAL: Correlation Alignment for Deep Domain Adaptation Baochen Sun and Kate Saenko University of Massachusetts Lowell, Boston University Abstract. Deep neural networks
More informationIMPROVING SPATIO-TEMPORAL FEATURE EXTRACTION TECHNIQUES AND THEIR APPLICATIONS IN ACTION CLASSIFICATION. Maral Mesmakhosroshahi, Joohee Kim
IMPROVING SPATIO-TEMPORAL FEATURE EXTRACTION TECHNIQUES AND THEIR APPLICATIONS IN ACTION CLASSIFICATION Maral Mesmakhosroshahi, Joohee Kim Department of Electrical and Computer Engineering Illinois Institute
More informationwith Deep Learning A Review of Person Re-identification Xi Li College of Computer Science, Zhejiang University
A Review of Person Re-identification with Deep Learning Xi Li College of Computer Science, Zhejiang University http://mypage.zju.edu.cn/xilics Email: xilizju@zju.edu.cn Person Re-identification Associate
More informationJoint Unsupervised Learning of Deep Representations and Image Clusters Supplementary materials
Joint Unsupervised Learning of Deep Representations and Image Clusters Supplementary materials Jianwei Yang, Devi Parikh, Dhruv Batra Virginia Tech {jw2yang, parikh, dbatra}@vt.edu Abstract This supplementary
More informationSpatial Latent Dirichlet Allocation
Spatial Latent Dirichlet Allocation Xiaogang Wang and Eric Grimson Computer Science and Computer Science and Artificial Intelligence Lab Massachusetts Tnstitute of Technology, Cambridge, MA, 02139, USA
More informationA Feature Selection Method to Handle Imbalanced Data in Text Classification
A Feature Selection Method to Handle Imbalanced Data in Text Classification Fengxiang Chang 1*, Jun Guo 1, Weiran Xu 1, Kejun Yao 2 1 School of Information and Communication Engineering Beijing University
More informationCRF Based Point Cloud Segmentation Jonathan Nation
CRF Based Point Cloud Segmentation Jonathan Nation jsnation@stanford.edu 1. INTRODUCTION The goal of the project is to use the recently proposed fully connected conditional random field (CRF) model to
More informationarxiv: v1 [cs.cv] 28 Nov 2018
CrowdCam: Dynamic Region Segmentation Nir Zarrabi Tel-Aviv University, Israel nirz@mail.tau.ac.il Shai Avidan Tel-Aviv University, Israel avidan@eng.tau.ac.il Yael Moses The Interdisciplinary Center, Israel
More informationTitle: Adaptive Region Merging Segmentation of Airborne Imagery for Roof Condition Assessment. Abstract:
Title: Adaptive Region Merging Segmentation of Airborne Imagery for Roof Condition Assessment Abstract: In order to perform residential roof condition assessment with very-high-resolution airborne imagery,
More informationA Keypoint Descriptor Inspired by Retinal Computation
A Keypoint Descriptor Inspired by Retinal Computation Bongsoo Suh, Sungjoon Choi, Han Lee Stanford University {bssuh,sungjoonchoi,hanlee}@stanford.edu Abstract. The main goal of our project is to implement
More informationSupplementary material: Strengthening the Effectiveness of Pedestrian Detection with Spatially Pooled Features
Supplementary material: Strengthening the Effectiveness of Pedestrian Detection with Spatially Pooled Features Sakrapee Paisitkriangkrai, Chunhua Shen, Anton van den Hengel The University of Adelaide,
More informationREJECTION-BASED CLASSIFICATION FOR ACTION RECOGNITION USING A SPATIO-TEMPORAL DICTIONARY. Stefen Chan Wai Tim, Michele Rombaut, Denis Pellerin
REJECTION-BASED CLASSIFICATION FOR ACTION RECOGNITION USING A SPATIO-TEMPORAL DICTIONARY Stefen Chan Wai Tim, Michele Rombaut, Denis Pellerin Univ. Grenoble Alpes, GIPSA-Lab, F-38000 Grenoble, France ABSTRACT
More informationGraph based machine learning with applications to media analytics
Graph based machine learning with applications to media analytics Lei Ding, PhD 9-1-2011 with collaborators at Outline Graph based machine learning Basic structures Algorithms Examples Applications in
More informationA Benchmark for Interactive Image Segmentation Algorithms
A Benchmark for Interactive Image Segmentation Algorithms Yibiao Zhao 1,3, Xiaohan Nie 2,3, Yanbiao Duan 2,3, Yaping Huang 1, Siwei Luo 1 1 Beijing Jiaotong University, 2 Beijing Institute of Technology,
More information