Citation Context

...odular functions has been used in sensor placement [6] and outbreak detection [9] problems. 1.1. Related Work Graph-based image segmentation work of Felzenszwalb and Huttenlocher (FH) [4], mean shift =-=[2]-=-, and watershed [23] are three of the most popular superpixel segmentation algorithms. FH and watershed are extremely fast; 2097mean shift is robust to local variations. However, they produce superpi...

Citation Context

...xperiments, it provides a speedup by a factor of 200–300 for image size 481x321 and on average requires 2.5 seconds. 5. Experiments We conducted the experiments on the Berkeley segmentation benchmark =-=[12]-=-. The benchmark contains 300 images with human-labeled ground truth segmentations. Superpixel segmentation has a different goal than object segmentation, therefore the performance metrics are also dif...

Citation Context

...mization of submodular functions has been used in sensor placement [6] and outbreak detection [9] problems. 1.1. Related Work Graph-based image segmentation work of Felzenszwalb and Huttenlocher (FH) =-=[4]-=-, mean shift [2], and watershed [23] are three of the most popular superpixel segmentation algorithms. FH and watershed are extremely fast; 2097mean shift is robust to local variations. However, they...

Citation Context

...nts. Then the pair M = (E,I) is a matroid. Maximization of a submodular function subject to a matroid constraint has been an active subject in combinatorial optimization; it is shown in Fisher et al. =-=[5]-=- that the greedy algorithm gives an 1 2 approximation bound. Following the same argument, we achieve the same ( 1 2 approximation) guarantee on the proposed greedy algorithm. A pseudocode is given in ...

Citation Context

...ploiting the matroid structure present in our formulation, we obtain a bound of 1 2 on the optimality of the solution. Recently, maximization of submodular functions has been used in sensor placement =-=[6]-=- and outbreak detection [9] problems. 1.1. Related Work Graph-based image segmentation work of Felzenszwalb and Huttenlocher (FH) [4], mean shift [2], and watershed [23] are three of the most popular ...

Citation Context

...ure present in our formulation, we obtain a bound of 1 2 on the optimality of the solution. Recently, maximization of submodular functions has been used in sensor placement [6] and outbreak detection =-=[9]-=- problems. 1.1. Related Work Graph-based image segmentation work of Felzenszwalb and Huttenlocher (FH) [4], mean shift [2], and watershed [23] are three of the most popular superpixel segmentation alg...

Citation Context

...he art in all the standard evaluation metrics. 1. Introduction Superpixel segmentation is an important module for many computer vision applications such as object recognition [15], image segmentation =-=[20, 8]-=-, and single view 3D reconstruction [7, 19]. A superpixel is commonly defined as a perceptually uniform region in the image. The major advantage of using superpixels is computational efficiency. A sup...

Citation Context

...he art in all the standard evaluation metrics. 1. Introduction Superpixel segmentation is an important module for many computer vision applications such as object recognition [15], image segmentation =-=[20, 8]-=-, and single view 3D reconstruction [7, 19]. A superpixel is commonly defined as a perceptually uniform region in the image. The major advantage of using superpixels is computational efficiency. A sup...

Citation Context

...utperforms the state of the art in all the standard evaluation metrics. 1. Introduction Superpixel segmentation is an important module for many computer vision applications such as object recognition =-=[15]-=-, image segmentation [20, 8], and single view 3D reconstruction [7, 19]. A superpixel is commonly defined as a perceptually uniform region in the image. The major advantage of using superpixels is com...

Citation Context

...to achieve a similar regularity. TurboPixel is based on evolving boundary curves from seeds uniformly placed in the image. Recently Veksler et al. [22] formulate superpixel segmentation as a GraphCut =-=[1]-=- problem. The regularity is enforced through a dense patch assignment technique for allowable pixel labels. These methods produce nice image tessellations as shown in [20, 10, 22]. Nevertheless, they ...

Citation Context

...atershed are extremely fast; 2097mean shift is robust to local variations. However, they produce superpixels with irregular sizes and shapes which tend to straddle multiple objects as pointed out in =-=[10, 22]-=-. Ren and Malik [20] propose using Normalized Cut (NCut) [21] for superpixel segmentation. NCut has the nice property of producing superpixels with similar sizes and compact shapes which is preferred ...

Citation Context

...cs. 1. Introduction Superpixel segmentation is an important module for many computer vision applications such as object recognition [15], image segmentation [20, 8], and single view 3D reconstruction =-=[7, 19]-=-. A superpixel is commonly defined as a perceptually uniform region in the image. The major advantage of using superpixels is computational efficiency. A superpixel representation greatly reduces the ...

Citation Context

...cs. 1. Introduction Superpixel segmentation is an important module for many computer vision applications such as object recognition [15], image segmentation [20, 8], and single view 3D reconstruction =-=[7, 19]-=-. A superpixel is commonly defined as a perceptually uniform region in the image. The major advantage of using superpixels is computational efficiency. A superpixel representation greatly reduces the ...

Citation Context

...atershed are extremely fast; 2097mean shift is robust to local variations. However, they produce superpixels with irregular sizes and shapes which tend to straddle multiple objects as pointed out in =-=[10, 22]-=-. Ren and Malik [20] propose using Normalized Cut (NCut) [21] for superpixel segmentation. NCut has the nice property of producing superpixels with similar sizes and compact shapes which is preferred ...

Citation Context

...undary recall as reported in [10, 22]. In contrast, our balancing objective, which regularizes the cluster sizes, avoids the over-smoothing problem and hence preserves object boundaries. Moore et al. =-=[14, 13]-=- propose an alternative approach for obtaining superpixels aligned with a grid. In [14], a greedy algorithm is used to sequentially cut images along some vertical and horizontal strips; whereas in [13...

Citation Context

...ifferent. We use three standard metrics which were commonly used for evaluating the quality of superpixels: undersegmentation error [10, 22], boundary recall [20] and achievable segmentation accuracy =-=[17]-=-. For the sake of completeness we first describe these metrics. We use G = {G1,G2,...,GnG} to represent a ground truth segmentation withnG segments and|Gi| denotes the segment size. • Undersegmentatio...

Citation Context

...undary recall as reported in [10, 22]. In contrast, our balancing objective, which regularizes the cluster sizes, avoids the over-smoothing problem and hence preserves object boundaries. Moore et al. =-=[14, 13]-=- propose an alternative approach for obtaining superpixels aligned with a grid. In [14], a greedy algorithm is used to sequentially cut images along some vertical and horizontal strips; whereas in [13...