Microsoft Technology Licensing, LLC

Qiang Huo

Lei Sun

Jun Du

Jian Qiao Sun

Peng Bai

Kai Chen

Zhengyou Zhang

Jinyu Li

Yifan Gong

Lei Ma

Jian Sun

Lijuan Wang

Frank Soong

Yongqiang Wang

Zhijie Yan

Zhi-Jie Yan

Wenping Hu

Chixiang Ma

Zhuoyao Zhong

Yuseok Ban

Implementations of the present disclosure provide a solution for text line detection. In this solution, a first text region comprising a first portion of at least a first text element and a second text region comprising a second portion of at least a second text element are determined from an image. A first feature representation is extracted from the first text region and a second feature representation is extracted from the second text region. The first and second feature representations comprise at least one of an image feature representation or a semantic feature representation of the image. A link relationship between the first and second text regions can then be determined based at least in part on the first and second feature representations. The link relationship can indicate whether the first and second portions of the first and second text elements are located in a same text line. In this way, by detecting text regions and determining the link relationship thereof based on their feature representations, the accuracy and efficiency for detecting text lines in various images can be improved

Text line detection

Some implementations disclosed herein provide techniques and arrangements to enable translating language characters in media content. For example, some implementations receive a user selection of a first portion of media content. Some implementations disclosed herein may, based on the first portion, identify a second portion of the media content. The second portion of the media content may include one or more first characters of a first language. Some implementations disclosed herein may create an image that includes the second portion of the media content and may send the image to a server. Some implementations disclosed herein may receive one or more second characters of a second language corresponding to a translation of the one or more first characters of the first language from the server.

Translating language characters in media content

According to implementations of the subject matter described herein, there is provided a solution for text recognition in an image. In this solution, a target text line area, which is expected to include a text to be recognized, is determined from an image. Probability distribution information of a character model element(s) present in the target text line area is determined using a single character model. The single character model is trained based on training text line areas and respective ground-truth texts in the training text line areas. Texts in the training text line areas are arranged in different orientations, and/or the ground-truth texts comprise texts are related to various languages (e.g., texts related to a Latin and an Eastern languages). The text in the target text line area can be determined based on the determined probability distribution information. The single character model enables more efficient and convenient text recognition.

Text recognition in image

The disclosure relates to a method and apparatus for training a learning machine, wherein the apparatus includes: a broadcasting module for broadcasting an initial global model for a training cycle to a plurality of worker nodes; a receiving module for receiving a plurality of updated local models from the plurality of worker nodes, wherein each updated local model is generated by one of the plurality of worker nodes independently based on a data split assigned to the worker node and the initial global model for the training cycle; an aggregating module for aggregating the plurality of updated local models to obtain an aggregated model; and a generating module for generating an updated global model for the training cycle based at least on the aggregated model and historical information which is obtained from a preceding training cycle.

Method and apparatus for training a learning machine

Some implementations may present a media file that includes video on a touchscreen display. A user gesture performed on the touchscreen display may be detected. The user gesture may include one of a tap gesture, a swipe gesture, or a tap and hold and drag while holding gesture. Text selected by the user gesture may be determined. One or more follow-up actions may be performed automatically based at least partly on the text selected by the user gesture.

Actionable content displayed on a touch screen

An image processing method and an apparatus () are provided, the method includes: obtaining a depth image of a protuberant object (); selecting a plurality of test points placed on a circle around a pixel in the depth image as a center point of the circle; calculating a protuberance value of the center point based on a comparison between the depth value of the center point and the depth value of each of the selected test points (); and determining one or more salient points of the protuberant object by using the protuberance value of each pixel in the depth image ().

Method and apparatus for detecting a salient point of a protuberant object

The use of the alternating direction method of multipliers (ADMM) algorithm to train a classifier may reduce the amount of classifier training time with little degradation in classifier accuracy. The training involves partitioning the training data for training the classifier into multiple data blocks. The partitions may preserve the joint distribution of input features and an output class of the training data. The training may further include performing an ADMM iteration on the multiple data blocks in an initial order using multiple worker nodes. Subsequently, the training of the classifier is determined to be completed if a stop criterion is satisfied following the ADMM iteration. Otherwise, if the stop criterion is determined to be unsatisfied following the ADMM iteration, one or more additional ADMM iterations may be performed on different orders of the multiple data blocks until the stop criterion is satisfied.

Deep learning using alternating direction method of multipliers

Dynamic texture mapping is used to create a photorealistic three dimensional animation of an individual with facial features synchronized with desired speech. Audiovisual data of an individual reading a known script is obtained and stored in an audio library and an image library. The audiovisual data is processed to extract feature vectors used to train a statistical model. An input audio feature vector corresponding to desired speech with which the animation will be synchronized is provided. The statistical model is used to generate a trajectory of visual feature vectors that corresponds to the input audio feature vector. These visual feature vectors are used to identify a matching image sequence from the image library. The resulting sequence of images, concatenated from the image library, provides a photorealistic image sequence with facial features, such as lip movements, synchronized with the desired speech. This image sequence is applied to the three-dimensional model.

Photo-realistic synthesis of three dimensional animation with facial features synchronized with speech

A high-dimensional posterior-based feature with partial distance elimination may be utilized for speech recognition. The log likelihood values of a large number of Gaussians are needed to generate the high-dimensional posterior feature. Gaussians with very small log likelihoods are associated with zero posterior values. Log likelihoods for Gaussians for a speech frame may be evaluated with a partial distance elimination method. If the partial distance of a Gaussian is already too small, the Gaussian will have a zero posterior value. The partial distance may be calculated by sequentially adding individual dimensions in a group of dimensions. The partial distance elimination occurs when less than all of the dimensions in the group are sequentially added.

Growing community of inventors

Beijing, China

Qiang Huo