Natasha Gelfand

Bio: Natasha Gelfand is an academic researcher from Nokia. The author has contributed to research in topics: Computational photography & Photography. The author has an hindex of 29, co-authored 40 publications receiving 4597 citations. Previous affiliations of Natasha Gelfand include Stanford University & Brown University.

Robust global registration

Abstract: We present an algorithm for the automatic alignment of two 3D shapes (data and model), without any assumptions about their initial positions. The algorithm computes for each surface point a descriptor based on local geometry that is robust to noise. A small number of feature points are automatically picked from the data shape according to the uniqueness of the descriptor value at the point. For each feature point on the data, we use the descriptor values of the model to find potential corresponding points. We then develop a fast branch-and-bound algorithm based on distance matrix comparisons to select the optimal correspondence set and bring the two shapes into a coarse alignment. The result of our alignment algorithm is used as the initialization to ICP (iterative closest point) and its variants for fine registration of the data to the model. Our algorithm can be used for matching shapes that overlap only over parts of their extent, for building models from partial range scans, as well as for simple symmetry detection, and for matching shapes undergoing articulated motion.

Outdoors augmented reality on mobile phone using loxel-based visual feature organization

Abstract: We have built an outdoors augmented reality system for mobile phones that matches camera-phone images against a large database of location-tagged images using a robust image retrieval algorithm. We avoid network latency by implementing the algorithm on the phone and deliver excellent performance by adapting a state-of-the-art image retrieval algorithm based on robust local descriptors. Matching is performed against a database of highly relevant features, which is continuously updated to reflect changes in the environment. We achieve fast updates and scalability by pruning of irrelevant features based on proximity to the user. By compressing and incrementally updating the features stored on the phone we make the system amenable to low-bandwidth wireless connections. We demonstrate system robustness on a dataset of location-tagged images and show a smart-phone implementation that achieves a high image matching rate while operating in near real-time.

Geometrically stable sampling for the ICP algorithm

Abstract: The iterative closest point (ICP) algorithm is a widely used method for aligning three-dimensional point sets. The quality of alignment obtained by this algorithm depends heavily on choosing good pairs of corresponding points in the two datasets. If too many points are chosen from featureless regions of the data, the algorithm converges slowly, finds the wrong pose, or even diverges, especially in the presence of noise or miscalibration in the input data. We describe a method for detecting uncertainty in pose, and we propose a point selection strategy for ICP that minimizes this uncertainty by choosing samples that constrain potentially unstable transformations.

Method, Device, Mobile Terminal, and Computer Program Product for a Point of Interest Based Scheme for Improving Mobile Visual Searching Functionalities

Abstract: Systems, methods, devices and computer program products which relate to utilizing a camera of a mobile terminal as a user interface for search applications and online services to perform visual searching are provided. The system consists of an apparatus that includes a processor that is configured to capture an image of one or more objects and analyze data of the image to identify an object(s) of the image. The processor is further configured to receive information that is associated with at least one object of the images and display the information that is associated with the image. In this regard, the apparatus is able to simplify access to location based services and improve a user's experience. The processor of the apparatus is configured to combine results of robust visual searches with online information resources to enhance location based services.

Reassembling fractured objects by geometric matching

Abstract: We present a system for automatic reassembly of broken 3D solids. Given as input 3D digital models of the broken fragments, we analyze the geometry of the fracture surfaces to find a globally consistent reconstruction of the original object. Our reconstruction pipeline consists of a graph-cuts based segmentation algorithm for identifying potential fracture surfaces, feature-based robust global registration for pairwise matching of fragments, and simultaneous constrained local registration of multiple fragments. We develop several new techniques in the area of geometry processing, including the novel integral invariants for computing multi-scale surface characteristics, registration based on forward search techniques and surface consistency, and a non-penetrating iterated closest point algorithm. We illustrate the performance of our algorithms on a number of real-world examples.

Guided Image Filtering

Abstract: In this paper, we propose a novel explicit image filter called guided filter. Derived from a local linear model, the guided filter computes the filtering output by considering the content of a guidance image, which can be the input image itself or another different image. The guided filter can be used as an edge-preserving smoothing operator like the popular bilateral filter [1], but it has better behaviors near edges. The guided filter is also a more generic concept beyond smoothing: It can transfer the structures of the guidance image to the filtering output, enabling new filtering applications like dehazing and guided feathering. Moreover, the guided filter naturally has a fast and nonapproximate linear time algorithm, regardless of the kernel size and the intensity range. Currently, it is one of the fastest edge-preserving filters. Experiments show that the guided filter is both effective and efficient in a great variety of computer vision and computer graphics applications, including edge-aware smoothing, detail enhancement, HDR compression, image matting/feathering, dehazing, joint upsampling, etc.

Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials

Abstract: Most state-of-the-art techniques for multi-class image segmentation and labeling use conditional random fields defined over pixels or image regions. While region-level models often feature dense pairwise connectivity, pixel-level models are considerably larger and have only permitted sparse graph structures. In this paper, we consider fully connected CRF models defined on the complete set of pixels in an image. The resulting graphs have billions of edges, making traditional inference algorithms impractical. Our main contribution is a highly efficient approximate inference algorithm for fully connected CRF models in which the pairwise edge potentials are defined by a linear combination of Gaussian kernels. Our experiments demonstrate that dense connectivity at the pixel level substantially improves segmentation and labeling accuracy.

Fast Point Feature Histograms (FPFH) for 3D registration

Abstract: In our recent work [1], [2], we proposed Point Feature Histograms (PFH) as robust multi-dimensional features which describe the local geometry around a point p for 3D point cloud datasets. In this paper, we modify their mathematical expressions and perform a rigorous analysis on their robustness and complexity for the problem of 3D registration for overlapping point cloud views. More concretely, we present several optimizations that reduce their computation times drastically by either caching previously computed values or by revising their theoretical formulations. The latter results in a new type of local features, called Fast Point Feature Histograms (FPFH), which retain most of the discriminative power of the PFH. Moreover, we propose an algorithm for the online computation of FPFH features for realtime applications. To validate our results we demonstrate their efficiency for 3D registration and propose a new sample consensus based method for bringing two datasets into the convergence basin of a local non-linear optimizer: SAC-IA (SAmple Consensus Initial Alignment).

Guided image filtering

Abstract: In this paper, we propose a novel type of explicit image filter - guided filter. Derived from a local linear model, the guided filter generates the filtering output by considering the content of a guidance image, which can be the input image itself or another different image. The guided filter can perform as an edge-preserving smoothing operator like the popular bilateral filter [1], but has better behavior near the edges. It also has a theoretical connection with the matting Laplacian matrix [2], so is a more generic concept than a smoothing operator and can better utilize the structures in the guidance image. Moreover, the guided filter has a fast and non-approximate linear-time algorithm, whose computational complexity is independent of the filtering kernel size. We demonstrate that the guided filter is both effective and efficient in a great variety of computer vision and computer graphics applications including noise reduction, detail smoothing/enhancement, HDR compression, image matting/feathering, haze removal, and joint upsampling.

Unsupervised Visual Representation Learning by Context Prediction

Abstract: This work explores the use of spatial context as a source of free and plentiful supervisory signal for training a rich visual representation. Given only a large, unlabeled image collection, we extract random pairs of patches from each image and train a convolutional neural net to predict the position of the second patch relative to the first. We argue that doing well on this task requires the model to learn to recognize objects and their parts. We demonstrate that the feature representation learned using this within-image context indeed captures visual similarity across images. For example, this representation allows us to perform unsupervised visual discovery of objects like cats, people, and even birds from the Pascal VOC 2011 detection dataset. Furthermore, we show that the learned ConvNet can be used in the R-CNN framework [19] and provides a significant boost over a randomly-initialized ConvNet, resulting in state-of-the-art performance among algorithms which use only Pascal-provided training set annotations.