Microsoft Technology Licensing, LLC

Simon A J Winder

Desney S Tan

Ashish Kapoor

James Anthony Fogarty

Richard S Szeliski

Gang Hua

Michael Kallay

Matthew Alun Brown

Charles L Zitnick

Jie Sun

A 'Point Cloud Smoother' provides various techniques for refining a 3D point cloud or other 3D input model to generate a smoothed and denoised 3D output model. Smoothing and denoising is achieved, in part, by robustly fitting planes to a neighborhood of points around each point of the input model and using those planes to estimate new points and corresponding normals of the 3D output model. These techniques are useful for a number of purposes, including, but not limited to, free viewpoint video (FVV), which, when combined with the smoothing techniques enabled by the Point Cloud Smoother, allows 3D data of videos or images to be denoised and then rendered and viewed from any desired viewpoint that is supported by the input data.

Smoothing and robust normal estimation for 3D point clouds

Various technologies described herein pertain to computing surface normals for points in a point cloud. The point cloud is representative of a measured surface of a physical object. A point in the point cloud can be set as a point of origin, and points in the point cloud can be modeled as electrostatic point charges. Moreover, a point of least electrostatic potential on a sphere centered at the point of origin can be computed as a function of the electrostatic point charges. Further, unit vector with a direction from the point of origin to the point of least electrostatic potential on the sphere can be assigned as a normal for the point of origin.

Surface normal computation on noisy sample of points

An interactive concept learning image search technique that allows end-users to quickly create their own rules for re-ranking images based on the image characteristics of the images. The image characteristics can include visual characteristics as well as semantic features or characteristics, or may include a combination of both. End-users can then rank or re-rank any current or future image search results according to their rule or rules. End-users provide examples of images each rule should match and examples of images the rule should reject. The technique learns the common image characteristics of the examples, and any current or future image search results can then be ranked or re-ranked according to the learned rules.

Interactive concept learning in image search

To render the comparison of image patches more efficient, the data of an image patch can be projected into a smaller-dimensioned subspace, resulting in a descriptor of the image patch. The projection into the descriptor subspace is known as a linear discriminant embedding, and can be performed with reference to a linear discriminant embedding matrix. The linear discriminant embedding matrix can be constructed from projection vectors that maximize those elements that are shared by matching image patches or that are used to distinguish non-matching image patches, while also minimizing those elements that are common to non-matching image patches or that distinguish matching image patches. The determination of such projection vectors can be limited such that only orthogonal vectors comprise the linear discriminant embedding matrix. The determination of the linear discriminant embedding matrix can likewise be constrained to avoid overfitting to training data.

Local image descriptors using linear discriminant embedding

A feature symbol triplets object instance recognizer and method for recognizing specific objects in a query image. Generally, the recognizer and method find repeatable features in the image, and match the repeatable features between a query image and a set of training images. More specifically, the recognizer and method finds features in the query image and then groups all possible combinations of three features in to feature triplets. Small regions or 'patches' in the query image, and an affine transformation is applied to the patches to identify any similarity between patches in a query image and training images. The affine transformation is computed using position of neighboring features in each feature triplet. Next, all similar patches are found, and then pairs of images are aligned to determine if the patches agree in the position of the object. If they do, then it is said that object is found and identified.

Growing community of inventors

Seattle, WA, United States of America

Simon A J Winder