Building a Non-Euclidean Roadmap from a Small Set of Images

By Silvina Rybnikov

This work addresses the problem of robotic navigation using natural visual features.

The system receives a set of target locations represented by images taken from those locations. The robot autonomously explores the environment, locates the targets, and builds a graph of the paths between them. The graph allows fast homing from any location in the environment to any target location.

Active Tracking and Pursuit Under Different Levels of Occlusions: a Two Layers Approach

By: Tomer Baum, Idan Izhaki, Ehud Rivlin, Gadi Katzir

We present an algorithm for a real-time, robust, vision-based active tracking and pursuit. The algorithm was designed to overcome problems emergent from active vision based pursuit, such as target occlusion. The algorithm offers to overcome an occlusion by two levels of reactions, we term layers. The purpose of the first layer is to cope with short term or medium term occlusions, i.e. occlusions where a known method such as Mean-Shift combined with a Kalman filter fails. For the first layer we designed the Hybrid filter for Active Pursuit (HAP). For long term occlusions we use the second layer. This layer is a decision algorithm following a learning procedure, and based on game theory related reinforcement .

A Segmentation Quality Measure Based On Rich Descriptors and Classification Methods

By: David Peles and Michael Lindenbaum

Most segmentation methods are based on a relatively simple score, designed to lend itself to relatively efficient optimization. We take the opposite approach and suggest more complex segmentation scores that are based on a mixture of on-line and off-line learning processes and rely on rich descriptors. The score is evaluated by a segmentation process which uses exploration-exploitation to search for good segments in various scales and shapes. We test our algorithm in a foreground-background segmentation task, given a minimal prior which is just a single seed point inside the object of interest. Results on two image databases are presented and compared with earlier approaches.

Non-Local Characterization of Scenery Images: Statistics, 3D Reasoning, and a Generative Model

By: Tamar Avraham, Michael Lindenbaum

This work focuses on characterizing scenery images. We semantically divide the objects in natural landscape scenes into background and foreground and show that the shapes of the regions associated with these two types are statistically different. We then focus on the background regions. We study statistical properties such as size and shape, location and relative location, the characteristics of the boundary curves and the correlation of the properties to the region's semantic identity. Then we discuss the imaging process of a simplified 3D scene model and show how it explains the empirical observations. We further show that the observed properties suffice to characterize the gist of scenery images, propose a generative parametric graphical model, and use it to learn and generate semantic sketches of new images, which indeed look like those associated with natural scenery.

Optimizing Gabor Filter Design for Texture Edge Detection and Classification

By: Roman Sandler and Michael Lindenbaum

An effective and efficient texture analysis method, based on a new criterion for designing Gabor filter sets, is proposed. The commonly used filter sets are usually designed for optimal signal representation.We propose here an alternative criterion for designing the filter set. We consider a set of filters and its response to pairs of harmonic signals.
Two signals are considered separable if the corresponding two sets of vector responses are disjoint in at least one of the components. We propose an algorithm for deriving the set of Gabor filters that maximizes the fraction of separable harmonic signal pairs in a given frequency range. The resulting filters differ significantly from the traditional ones. We test these maximal harmonic discrimination (MHD) filters in several texture analysis tasks: clustering, recognition, and edge detection. It turns out that the proposed filters perform much better than the traditional ones in these tasks. They can achieve performance similar to that of state-of-the-art, distribution based (texton) methods, while being simpler and more computationally efficient.

Unsupervised estimation of segmentation quality using nonnegative factorization

By: Roman Sandler and Michael Lindenbaum

We propose an unsupervised method for evaluating image segmentation. Prevalent methods are typically based on valuating smoothness within segments and contrast between them. The proposed approach differs: it provides a meaningful, uantitative assessment of segmentation quality in precision/recall terms, applicable until now only for supervised evaluation. oreover, it builds on a new image model, which characterizes the segments as a mixture of basic feature distributions. The istributions are obtained by a nonnegative factorization (NMF) process and precision/recall estimates are then estimated from hem. As the estimates are based on the intrinsic properties of the image being evaluated and not on a comparison to typical mages (learning), they are relatively robust to context factors such as image quality or texture. Experimental results demonstrate he accuracy of the precision/recall estimates in comparison to human-judged ground truth. Finally, the unsupervised measure an be used to tune and improve the quality of popular segmentation algorithms.

Nonnegative matrix factorization with Earth Mover’s Distance Metric

By: Roman Sandler and Michael Lindenbaum

Nonnegative Matrix Factorization (NMF) approximates a given data matrix as a product of two low rank nonnegative matrices, usually by minimizing the L2 or the KL distance between the data matrix and the matrix product. This factorization was shown to be useful for several important computer vision applications.
We propose here a new NMF algorithm that minimizes the Earth Mover’s Distance (EMD) error between the data and the matrix product. We propose an iterative NMF algorithm (EMD NMF) and prove its convergence. The algorithm is based on linear programming. We discuss the numerical difficulties of the EMD NMF and propose an efficient approximation.
Naturally, the matrices obtained with EMD NMF are different from those obtained with L2 NMF. We discuss these differences in the context of two challenging computer vision tasks – texture classification and face recognition – and demonstrate the advantages of the proposed method.

A statistically correct estimation of epipolar geometry

By: Stas Rozenfeld, Ilan Shimshoni and Michael Lindenbaum

The fundamental matrix is an essential tool for characterizing the relative geometry of two cameras. The matrix should be consistent with the image data and, at the same time, satisfy a singularity constraint. The fundamental matrix is usually estimated by means of an initial solution, which is calculated from the image data and then modified to satisfy the singularity constraint, by zeroing its smallest singular value. This approach, however, produces suboptimal results, especially when the amount of image data is small. This is the case when the matrix is estimated within a RANSAC process. We argue that this deficiency is due to (implicit) incorrect statistical modeling, which we rectify in this paper. The proposed method propagates the image noise distribution to a distribution on the initial solutions. It then uses this distribution to find a maximum likelihood solution which satisfies the constraint. The possibility of such a method has been considered in the literature, but was assumed to be inefficient in the RANSAC context. The same general technique can be applied to produce an improved estimate of a large class of models. In this work, it was also applied to estimating the essential matrix, which must satisfy an additional constraint. Our experiments show greater accuracy, at a modest computational cost. When used in the context of a RANSAC procedure, this accuracy enhancement leads to faster algorithms.

Visual Tracking Of Object Silhouettes

By: Guy Boudoukh, Ido Leichter and Ehud Rivlin

To appear in 2009 IEEE International Conference on Image Processing

In this research we propose a new method that addresses the problem of visually tracking the bitmap (silhouette) of an object in a video under very general conditions. We assume a general target, possibly non rigid, with no prior information except initialization. The target, as well as the background, may change its appearance over time and the camera may move arbitrarily. The proposed algorithm fuses different visual cues by means of a conditional random field probabilistic model. The target's bitmap is estimated every frame by incorporating temporal color similarity, spatial color continuity and spatial motion continuity into an energy function that is minimized via min-cut. The spatial motion continuity is incorporated in the energy function in multiple image resolutions by a novel multi-scale energy term. Compared to other methods that calculate optical flow for the whole image, the algorithm complexity is reduced when the optical flow calculation is done only at specific feature points. Experimental results show that our method outperforms other algorithms that address the problem of tracking under general conditions. Experiments on a variety of video clips demonstrate the robustness and effectiveness of our method to track an object under very general conditions.

Vesicles & Amoebae Or On Globally Constrained Shape Deformations

By: I. Goldin,   J. M. Delosme,   A. M. Bruckstein

Modeling the deformation of shapes under constraints on both perimeter and area is a challenging task due to the highly nontrivial interaction between the need for °exible local rules for manipulating the boundary and the global constraints. We propose several methods to address this problem and generate "random walks" in the space of shapes obeying quite general possibly time varying constraints on their perimeter and area. Design of perimeter and area preserving deformations are an interesting and useful special case of this problem. The resulting deformation models are employed in annealing processes that evolve original shapes toward shapes that are optimal in terms of boundary bending-energy or other functionals. Furthermore, such models may ¯nd applications in the analysis of sequences of real images of deforming objects obeying global constraints as building blocks for registration and tracking algorithms.