167 results about "Text annotation" patented technology

Streaming compressed digital hypervideo received upon a digital communications network is decoded (decompressed) and played in a client-computer-based "video on web VCR" software system. Scene changes, if not previously marked upstream, are automatically detected, and typically twenty-one past scenes are displayed as thumbnail images. Hyperlinks within the main video scene, and / or any thumbnail image, show as hotspots, with text annotations typically appearing upon a cursor "mouse over". All hyperlinks-as are provided and inserted by, inter alia, the upstream network service provider (the "ISP")-may be, and preferably are, full-custom dynamically-resolved to each subscriber / user / viewer ("SUV") upon volitional "click throughs" by the SUV, including retrospectively on past hypervideo scenes as appear within the thumbnail images. Hyperlinking permits (i) retrieving information and commercials, including streaming video / hypervideo, from any of local storage, a network (or Internet) service provider ("ISP"), a network content provider, and / or an advertiser network site, (ii) entering a contest of skill or a lottery of chance, (iii) gambling, (iv) buying (and less often, selling), (v) responding to a survey, and expressing an opinion, and / or (vi) sounding an alert.

A fact extraction tool set (“FEX”) finds and extracts targeted pieces of information from text using linguistic and pattern matching technologies, and in particular, text annotation and fact extraction. Text annotation tools break a text, such as a document, into its base tokens and annotate those tokens or patterns of tokens with orthographic, syntactic, semantic, pragmatic and other attributes. A user-defined “Annotation Configuration” controls which annotation tools are used in a given application. XML is used as the basis for representing the annotated text. A tag uncrossing tool resolves conflicting (crossed) annotation boundaries in an annotated text to produce well-formed XML from the results of the individual annotators. The fact extraction tool is a pattern matching language which is used to write scripts that find and match patterns of attributes that correspond to targeted pieces of information in the text, and extract that information.

A method and apparatus for video retrieval and cueing that automatically detects human faces in the video and identifies face-specific video frames so as to allow retrieval and viewing of person-specific video segments. In one embodiment, the method locates human faces in the video, stores the time stamps associated with each face, displays a single image associated with each face, matches each face against a database, computes face locations with respect to a common 3D coordinate system, and provides a means of displaying: 1) information retrieved from the database associated with a selected person or people, 2) path of travel associated with a selected person or people 3) interaction graph of people in video, 4) video segments associated with each person and / or face. The method may also provide the ability to input and store text annotations associated with each person, face, and video segment, and the ability to enroll and remove people from database. The videos of non-human objects may be processed in a similar manner. Because of the rules governing abstracts, this abstract should not be used to construe the claims.

An improved command handler (and database system utilizing the improved command handler) interfaces to a datastore storing item data for a plurality of items and bi-directional modifier data, corresponding to a relation between at least one first item and at least one second item, that represents first text characterizing semantics of a relationship of the at least one first item to the at least one second item, and represents second text characterizing semantics of a relationship of the at least one second item to the at least one first item. The command handler operates, in response to receiving a first-type query command that specifies at least one given item, to access the datastore to identify i) at least one related item that is related to the given item, and identify ii) either the first text or the second text characterizing semantics of the relation between the given item and the at least one related item. The command hander returns i) data corresponding to the at least one related item; and ii) data corresponding to the identified first text and or second text characterizing semantics of the relation between the given item and the at least one related item. Preferably, the data returned in response to the first-type query command identifies the at least one related item. and identifies the first text or second text characterizing semantics of the relation between the given item and the at least one related item. In addition, the command handler preferably supports additional commands that retrieve from the datastore information related to specified objects, object types, and relations.The command handler (and database system) of the present invention may be used in a wide assortment of software applications, including enterprise applications (such as e-business applications, supply chain management applications, customer relationship management applications, decision support applications), the file system in operating systems, web browsers, e-mail applications and personal information management applications. Importantly, the command handler (and database system) provides an efficient mechanism to query the organization of the data elements (and the relationships therebetween) stored and accessed in such software applications, in a manner that efficient and readily adaptable to client-server database systems or other distributed database systems.

To facilitate the use of audio files for annotation purposes, an audio file format, which includes audio data for playback purposes, is augmented with a parallel data channel of line identifiers, or with a map associating time codes for the audio data with line numbers on the original document. The line number-time code information in the audio file is used to navigate within the audio file, and also to associate bookmark links and captured audio annotation files with line numbers of the original text document. An annotation device may provide an output document wherein links to audio and / or text annotation files are embedded at corresponding line numbers. Also, a navigation index may be generated, having links to annotation files and associated document line numbers, as well as bookmark links to selected document line numbers.

A media diary or media management application implemented in a digital device that incorporates a bookmarking and / or annotating mechanism for providing graphical enhancement to media files, calendar events and time periods displayed in the media management application. In addition, the bookmarking and annotation mechanism provides for text notes to be associated with the media files, calendar event or time period and for the text note to be accessible through the bookmark or annotation.

A method (and corresponding database system) for displaying in a view window information characterizing semantics of relations between objects. For each given relation between at least one subject object and at least one direct object, bi-directional modifier data is stored that represents first text characterizing semantics of a relationship of the at least one first object to the at least one second object, and represents second text characterizing semantics of a relationship of the at least one second object to the at least one first object. In response to predetermined user input associated with an object node displayed in the view window, a set of relations whose at least one subject object or at least one direct object is associated with the object node is identified. For at least one relation in the set of relations, the view window is updated to include a second node comprising a graphical representation of: the first text of the given relation in the event that the given object is a subject object in the given relation, or the second text of the given relation in the event that the given object is a direct object in the given relation. The second node may be a relation node associated with a given relation, or a mixed node associated with a relation-type pair. In response to predetermined input with a second node, the second node may be expanded to identify and display one or more object nodes (identifying direct object(s) of relations derived from expansion of a subject object associated therewith or identifying subject object(s) of relations derived from expansion of a direct object associated therewith). Preferably, this expansion routine is recursive in nature.The method (and database system) of the present invention may be used in a wide assortment of software applications, including enterprise applications (such as e-business applications, supply chain management applications, customer relationship management applications, decision support applications), the file system in operating systems, web browsers, e-mail applications and personal information management applications. Importantly, the method (and database system) provides an easy, user friendly and efficient mechanism to define, view and query the organization of the data elements (and the relationships therebetween) stored and accessed in such software applications, in a manner that conveys the real-world meaning of such relationships.

The invention discloses a named entities recognition method based on bidirectional LSTM and CRF. The named entities recognition method based on the bidirectional LSTM and CRF is improved and optimizedbased on the traditional named entities recognition algorithm in the prior art. The named entities recognition method based on the bidirectional LSTM and CRF comprises the following steps: (1) preprocessing a text, extracting phrase information and character information of the text; (2) coding the text character information by means of the bidirectional LSTM neural network to convert the text character information into character vectors; (3) using the glove model to code the text phrase information into word vectors; (4) combining the character vectors and the word vectors into a context information vector and putting the context information vector into the bidirectional LSTM neural network; and (5) decoding the output of the bidirectional LSTM with a linear chain condition random field to obtain a text annotation entity. The invention uses a deep neural network to extract text features and decodes the textual features with the condition random field, therefore, the text feature information can be effectively extracted and good effects can be achieved in the entity recognition tasks of different languages.

An e-book allowing note and corrigendum sharing as well as differential update is produced using vector graphics satisfying XML syntax, and a script program is embedded in or linked to the e-book. The script program allows a user to add various kinds of note markups to the e-book, including fluorescent marker, text annotation, handwriting note, bookmark, digital blackboard record, post-it note, highlight, voice, memo, tracking information, etc. The e-book also allows an e-book editor to correct any error in the e-book. The note markups and corrigenda may be uploaded via network, and be shared among other users by way of differential update.

The invention discloses a training method for multiple personalized acoustic models, a voice synthesis method and a voice synthesis device. The training method comprises the following steps: training a reference acoustic model according to first acoustic feature data of training voice data and first text annotation data corresponding to the training voice data; acquiring voice data of a target user; training a first target user acoustic model according to the reference acoustic model and the voice data; generating second acoustic feature data of the first text annotation data according to the first target user acoustic model and the first text annotation data; training a second target user acoustic model according to the first text annotation data and the second acoustic feature data. According to the model training method disclosed by the embodiment, in a process of training a target user acoustic model, the requirement on the scale of voice data of the target user is lowered, and a plurality of personalized acoustic models including the voice features of the target user can be trained by using a small amount of user voice data.

Online educational videos are often difficult to navigate. Furthermore, most video interfaces do not lend themselves to note-taking. Described system detects and reuses boundaries that tend to occur in these types of videos. In particular, many educational videos are organized around distinct breaks that correspond to slide changes, scroll events, or a combination of both. Described algorithms can detect these structural changes in the video content. From these events the system can generate navigable overviews to help users searching for specific content. Furthermore, these boundary events can help the system automatically associate rich media annotations to manually-defined bookmarks. Finally, when manual or automatically recovered spoken transcripts are available, the spoken text can be combined with the temporal segmentation implied by detected events for video indexing and retrieval. This text can also be used to seed a set of text annotations for user selection or be combined with user text input.

A method and system for displaying information characterizing the semantics of relations between objects. For each given relation between at least one subject object and at least one direct object, bi-directional modifier data is stored in a separate table and characterizes the semantics of a relationship between objects of different type. A bi-directional modifier is stored in distinct tables and comprises arbitrary text strings defined by user input. Test is linked to a relation between at two different object types and represents text. The present invention may be used in a wide assortment of software applications, including enterprise applications.

One or more audio / video files are provided on a central server (203), accessible via a computer network. A audio / video file is requested by a user (205) and the file is transmitted to that user for viewing (207). The user enters edit point information specifying a portion of the previously-transmitted audio / video file relating to which the user wishes to make an annotation. The edit point information is received (209) by the central server over the computer network along with the textual annotation entered by the user. Use may be made of an optional rule that the edit point information must satisfy before it is accepted (211). The received edit point information and textual annotation are stored in an annotation data file (213). A subsequent user may request the annotation data file, which is transmitted to that user. The annotation text along with the relevant portion of the audio / vide file is then displayed for the requesting user.

The invention discloses a method for image automatic text annotation based on generative adversarial networks, which comprises the steps of generating false statements by a generator, rebuilding a discriminator at the same time, and inputting the generated statements and real statements into the discriminator to train until the discriminator cannot discriminate the real statements and the generated statements. The method disclosed by the invention changes a problem that sentences are rigid and inflexible in CNN-RNN image automatic statement annotation, and enables the generated sentences to be more accurate, more natural and more diversified. The generated statements can face more complex scenes in reality and more conform to language expression means annotation images of the human, thereby having more extensive applications in practice.

The invention discloses a text annotation method and a playing method of video files and corresponding systems thereof, wherein, the text annotation method of the video files comprises the step of carrying out annotation by lens on the video files by utilizing the information of the script mark number and the script mark sheet and the like in film and television pieces. The quick search of the video contents can be realized by searching for the self defining keywords in the text annotation information. Based on the search of text annotation information, the playing method of video files realizes that users can acquire corresponding contents in the video according to individual preference and requirements; specific contents in the video files are played by reorganizing corresponding contents, thus providing viewing experience of 'self defining' playing video for users.

A system and method for capturing, displaying, and navigating text annotations in a non-modifiable document is disclosed. Once it is determined that a text annotation is to be created, the system determines the file position of the selected object. The file position of the selected object is stored along with the created text annotation in another file or a non-read only portion of a file storing the document. Using the file position, the text annotation may be properly identified with the selected object without modifying the non-modifiable document. Once a text annotation is displayed a user may easily navigate among the captured annotations.

A system and method for annotating three-dimensional electronic documents. A user specifies, on a two-dimensional screen, a portion of a page of a three-dimensional document as a specific page area to be annotated. The annotation is displayed to the user with the transparent polygon technique, the vertex coloring technique and / or the texture coloring technique. These display techniques support highlighting annotations, free-form annotations, and text annotations.

The invention discloses content-based data retrieval methods for unmanned aerial vehicle spying images, belonging to the technical field of image retrieval. Two content-based retrieval methods for the unmanned aerial vehicle spying images are realized in a data management system; in the first content-based retrieval method, sorting is carried out according to a Hamming distance of fingerprint sequences of two images on the basis of a perception hash algorithm for searching images with images; and, in the second content-based retrieval method, sorting is carried out according to a colour distribution similarity value of the two images by selecting colour characteristics of the unmanned aerial vehicle spying images. According to the invention, information retrieval can be directly extracted from contents of the unmanned aerial vehicle spying images; related text annotations of the images are unnecessary; characteristic extracting and index establishing in the invention can be automatically realized through a computer; therefore, the retrieval efficiency is greatly improved; the content-based data retrieval methods disclosed by the invention are capable of searching related images in a large unmanned aerial vehicle image data management system at a rapid speed; objects in the unmanned aerial vehicle spying images are unnecessary to understand and identify; and thus, quick content-based retrieval is concerned.

The present invention discloses a textual annotation system for video recordings. Such a system can include a video recording, an annotation interface, and an annotation manager. The annotation interface can be configured to accept and execute a set of user-selectable commands for creating, modifying, and transmitting a user-created annotation set. The user-created annotation set can contains multiple textual annotations with associated placement data. The annotation manager can be configured to display the user-created annotation set synchronously with the video recording such that the contents of the video recording are unaltered.

A text annotation structured storage system stores text annotations with associated type information in a structured data store. The present system persists or stores annotations in a structured data store in an indexable and queryable format. Exemplary structured data stores comprise XML databases and relational databases. The system exploits type information in a type system to develop corresponding schemas in a structured data model. The system comprises techniques for mapping annotations to an XML data model and a relational data model. The system captures various features of the type system, such as complex types and inheritance, in the schema for the persistent store. In particular, the repository provides support for path navigation over the hierarchical type system starting at any type.

The invention provides a method for training an acoustic model based on CTC (Connectionist Temporal Classification). The method comprises the steps of 1, training an initial GMM (Gaussian Mixture Model), wherein time point forced alignment on text annotation of training data by using the GMM to obtain a time region corresponding to each phoneme; 2, inserting a blank symbol associated with the phoneme behind each phoneme, wherein each phoneme has a unique blank symbol; 3, constructing a CTC forward and backward calculated search path diagram for a phoneme annotation sequence with the blank symbols being added by adopting a finite state machine; 4, restricting the appearance time range of each phoneme according to a time alignment result, pruning the search path diagram, and cutting off thepath with the phoneme position exceeding the time restrictions so as to obtain a final search path diagram required by calculating a network error in CTC; and 5, performing acoustic model training byadopting the combination of a time-delay neural network (TDNN) structure and the CTC method to obtain a final TDNN-CTC acoustic model.

The invention relates to a text detection model training method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a sample image with a text annotation box; performing feature extraction on the sample image, inputting the sample feature map into an initial text detection model to obtain a text prediction box of the sample image and a corresponding second coordinate matrix, and determining a loss value of the second coordinate matrix corresponding to the text prediction box according to a preset loss function and the first coordinate matrixof the text annotation box; and training an initial text detection model according to the loss value. A polygonal text annotation box with any angle is arranged in a sample image used during training; therefore, the directivity and the scale change of the text can be learned in the process of training the model, and when the model is used for text positioning detection, the target textbox which is matched with the text direction in the image and has any angle can be output, so that the recognition precision of a subsequent machine is improved.

The invention discloses a sensitive data discovery method based on text recognition. The sensitive data discovery method comprises the following steps of S01, extracting the sample data; S02, constructing a training sample, collecting a text data set, and constructing the training sample; S03, training a sample annotation model, obtaining a training sample based on S02, and training a text annotation model; S04, constructing data features; S05, constructing a training set, carrying out label description on the data set obtained in the S04 to form a training set for constructing a classification judgment model; S06, constructing a classification judgment model, and forming a variable prediction model according to the training set obtained in the S05; S07, testing the model. Through the identification of the data variables, the sensitive data can be accurately and efficiently judged and identified under the condition that the data dictionary and the matching rules are incomplete, and theconsistency of identification and classification results is ensured.

The invention belonging to the field of automatic speech recognition technology relates to an automatic speech recognition method based on a random depth delay neural network model. The method comprises: preparing training data; extracting acoustic features from trained speech audio data; training a traditional GMM-HMM model and carrying out forced alignment on the trained speech audio data by using the trained GMM-HMM model to obtain a corresponding frame level training label; supervising and training a random-depth-based time-delay neural network model by using the trained speech audio dataand the corresponding frame level training label and acquiring an acoustic model by combining a hidden Markov model; carrying out training by using corresponding text annotation data or texts of otherdata sets to obtain a trained language model; and constructing an automatic speech recognition decoder by using the trained language model and acoustic model. Therefore, the modeling ability of the model is strengthened and problems of over-fitting and gradient disappearing during the training process are solved, so that the accuracy of the speech recognition is improved.

Disclosed is a semi-supervised self-learning driven medical text disease identification method provided by the invention, feature classification of the medical text is mainly realized, the features include disease object types, disease progression, whether diseases occur, disease severity, disease condition, disease uncertainty and the like, and by identifying and classifying features, unstructured medical texts are structured and can be directly processed and used in further information mining; based on the features of few medical text annotation texts and many unannotated texts, the invention is developed from the aspects of feature extraction and classification model optimization; experimental results show that the method well makes up for the defects and deficiencies of few tagged texts, and since similar data and wrong classification data are easily introduced into the self-learning method, the effect of the semi-supervised SVM in the aspect of utilizing untagged data is better than that of the semi-supervised SVM, and the feasibility and high efficiency of the method are proved.