728 results about "Text graph" patented technology

A method for presenting advertising in an interactive service provided on a computer network, the service featuring applications which include pre-created, interactive text / graphic sessions is described. The method features steps for presenting advertising concurrently with service applications at the user terminal configured as a reception system. In accordance with the method, the advertising is structured in a manner comparable to the service applications enabling the applications to be presented at a first portion of a display associated with the reception system and the advertising presented at a second portion. Further, steps are provided for storing and managing advertising at the user reception system so that advertising can be pre-fetched from the network and staged in anticipation of being called for presentation. This minimizes the potential for communication line interference between application and advertising traffic and makes the advertising available at the reception system so as not to delay presentation of the service applications. Yet further the method features steps for individualizing the advertising supplied to enhance potential user interest by providing advertising based on a characterization of the user as defined by the users interaction with the service, user demographics and geographical location. Yet additionally, advertising is provided with transactional facilities so that users can interact with it.

A method for generating an index of the text of a video image sequence is provided. The method includes the steps of determining the image text objects in each of a plurality of frames of the video image sequence; comparing the image text objects in each of the plurality of frames of the video image sequence to obtain a record of frame sequences having matching image text objects; extracting the content for each of the similar image text objects in text string format; and storing the text string for each image text object as a video text object in a retrievable medium.

An apparatus and method facilitate combined query based searching with serendipitous browsing in a multimedia collection. A user selects objects to label from a local map, which may include representations of objects retrieved from the collection as being responsive to a text or image base query. The text and image portions of the object can be independently labeled. Unlabeled objects are scored and ranked based on the applied labels of labeled objects, which may take into account cross-media pseudo-relevance and user selectable (or default) parameters, such as a forgetting factor, which tends to place greater weight on more recently labeled objects, and a modality parameter, which laces greater weight on the modality (text, image, or hybrid) currently selected by the user. The local map is modified, based on the ranking, optionally after reranking of objects to improve the diversity of the displayed objects.

A method and apparatus for organizing and processing interconnected pieces of information (“nodes”) using a digital computer is disclosed. Each node has elements that may be text, images, audio, video, and other computer programs. A graph-based user interface presents the individual nodes in spatial arrangements that reflect the relationships among the nodes. User interaction indicating interest in a particular node results in an increase in the “activation” of that node. This leads to an increase in the size of the presentation of that node, as well as an increase in the size of the presentation of closely related nodes. The result is a unique user interaction paradigm that allows for intuitive traversal of complex collections of nodes.

The invention belongs to the technical field of image processing, and specifically relates to an English text detection method with a text direction correction function. The method comprises the following steps of: respectively carrying out maximally stable extremal region detection on each channel of an English text image so as to obtain candidate text regions; establishing a convolutional neuralnetwork model-based classifier, and filtering wrong candidate text regions so as to obtain preliminary text regions; grouping the preliminary text regions by utilizing a double-layer text grouping algorithm; and carrying out direction correction on the grouped preliminary text regions so as to obtain a corrected text. According to the method, an enhanced multichannel MSER model is utilized to obtain more refined text regions, a parallel SPP-CNN classifier is imported to better distinguish text regions and non-text regions, images with any sizes can be processed, and pool features can be extracted under multiple scales, so that more features can be known through multilayer space information of source images; and the method is capable of processing slightly inclined scene texts.

The invention discloses a character recognition method based on an attention mechanism and linkage time classification loss. The character recognition method comprises the following steps that S1, a data set is collected; s2, carrying out preprocessing such as scale zooming, gray level conversion and pixel normalization on the image sample; s3; processing the tag sequence of the sample, includingfilling, coding and word embedding; s4, constructing a convolutional neural network, and performing feature extraction on the text image processed in the step S3; s5, encoding the features extracted in the step S4 by using a stacked bidirectional recurrent neural network to obtain encoded features; s6, inputting the coding features obtained in the S5 into a connection time classification model tocalculate a prediction probability; and S7, calculating weights of different coding characteristics by using an attention mechanism to obtain coded semantic vectors.

The invention discloses a multi-modal knowledge graph construction method, and relates to the knowledge engineering technology in the field of big data. The method is realized through the following technical scheme: firstly, extracting multi-modal data semantic features based on a multi-modal data feature representation model, constructing a pre-training model-based data feature extraction model for texts, images, audios, videos and the like, and respectively finishing single-modal data semantic feature extraction; secondly, projecting different types of data into the same vector space for representation on the basis of unsupervised graph, attribute graph, heterogeneous graph embedding and other modes, so as to realize cross-modal multi-modal knowledge representation; on the basis of the above work, two maps needing to be fused and aligned are converted into vector representation forms respectively, then based on the obtained multi-modal knowledge representation, the mapping relation of entity pairs between knowledge maps is learned according to priori alignment data, multi-modal knowledge fusion disambiguation is completed, decoding and mapping to corresponding nodes in the knowledge maps are completed, and a fused new atlas, entities and attributes thereof are generated.

A system and method for computing confidence in an output of a text recognition system includes performing character recognition on an input text image with a text recognition system to generate a candidate string of characters. A first representation is generated, based on the candidate string of characters, and a second representation is generated based on the input text image. A confidence in the candidate string of characters is computed based on a computed similarity between the first and second representations in a common embedding space.

The invention provides an image compression method and equipment, and a system. The method comprises: according to a preset image block size, segmenting a to-be-compressed image into a plurality of image blocks that are not overlapped mutually; according to the colors and / or texture features of all image blocks, determining types of all the image blocks; and on the basis of a mapping relation between a preconfigured image type and a compression rule as well as the types of all the image blocks, carrying out compression coding on all image blocks so as to complete compression of the to-be-compressed image. According to the method, the equipment, and the system, effective compression of a mixed image containing text graph information and natural image information can be realized.

The invention relates to the field of automatic abstracting, and discloses a graph model-based automatic abstracting method. According to the technical scheme, an LDA probability topic model is applied to measurement of semantic correlation between sentences and improvement of the measurement effect of sentence correlation; and an idea of topic correlation and position sensitivity of the sentences is provided, so that abstract generation is relatively reasonable and effective. The method comprises the following steps: firstly, obtaining topic probability distribution of a document and word probability distribution of the topic through training the LDA topic model, determining the topic probability distribution of the sentences and effectively converting a semantic similarity measurement between the sentences into a similarity measurement problem of the topic probability distribution of the sentences; with the sentences as nodes, building edges by referring tothe cosine similarity and according to the semantic similarity between the sentences and generating a text graph representing the document; calculating the topic correlation between the sentences according to the topic probability distribution of the sentences and the topic probability distribution of the document; and calculating the position sensitivity and the like of the sentences according to the positions of the sentences in the document.

The invention discloses a method for correcting a text image and belongs to the field of optical character recognition. The method comprises the following steps of: positioning a character zone in the text image and extracting suspected characters; recognizing the suspected characters, wherein if the recognition reliability is higher than the reliability reference value of a single character, the effective direction of the characters is positive, otherwise, the suspected characters are rotated counterclockwise respectively for judging the effective direction; extracting and sending the characters having the effective direction to an effective character set; extracting the corresponding text direction until the accumulated value of the recognition reliability in a certain direction is higher than the predetermined recognition reliability; correcting the direction of the image and recognizing and outputting the text image. The method effectively eliminates the influence from unreliable characters, prevents the unreliable zone in the character zone from influencing the judgment of the character direction, is well suitable for the text images having complexity or noise or poor quality and can judge the direction of a text image fast and correctly so as to effectively recognize the text image.

The invention discloses a scene text recognition method based on sparse coding characteristics, and relates to computer vision and pattern recognition. The method includes the steps: inputting a natural scene text image to be recognized; by the aid of a multi-scale sliding window method, detecting and recognizing a window area in the image by a character classifier, for each character class, determining a large output area of the classifier as a candidate character area, determining a small output area as a background area, finding the candidate character area in the image, retaining the area with the largest output value of the classifier and the corresponding character class for the area with a large overlapping ratio by the aid of a non-maximum suppression method, and removing the repetitive and redundant candidate character area to obtain a character detection result; combining detected characters into a word or text line; outputting a scene text recognition result. Structural characteristics of the characters can be more effectively expressed and extracted, so that the recognition rate of a scene text is increased.

A method and a system of multi-directional scene text recognition based on multi-element attention mechanism are provided. The method includes: performing normalization processing for a text row / column image I output from an external text detection module by a feature extractor, extracting a feature for the normalized image by using a deep convolutional neural network to acquire an initial feature map F0, and adding a 2-dimensional directional positional encoding P to an initial feature map F0 in order to output a multi-channel feature map F; converting the multi-channel feature map F output from a feature extractor by an encoder into a hidden representation H; and converting the hidden representation H output from the encoder into a recognized text by a decoder and using the recognized text as the output result. The method and the system of multi-directional scene text recognition based on multi-element attention mechanism provided by the present invention are applied to multi-oriented scene text images including horizontal text, vertical text, and curved text etc., and have achieved high applicability.

The invention relates to a curly text image preprocessing method and a lottery ticket scanning recognition method. The curly text image preprocessing method comprises the steps of carrying out gray stretch on a text image and enhancing the edge effect, carrying out binaryzation on the text image after gray stretch, carrying out edge extraction on the text image after binaryzation, carrying out image rotation and correction on the text image according to extracted edge, and straightening and correcting the curly edge. The lottery ticket scanning recognition method is based on the curly text image preprocessing method, and comprises the steps that curly text image preprocessing is carried out on a lottery ticket after the lottery ticket is scanned, then character information on the preprocessed and scanned lottery ticket is recognized through the OCR engine recognition technology, and further whether a prize is won in a lottery is determined. The curly text image preprocessing method and the lottery ticket scanning recognition method improve the recognition accuracy rate of lottery ticket class curly text images and have reference significance for recognition of the curly text images under similar unsatisfactory conditions.

A method for learning parameters used to recognize characters included in a text in a scene text image of training set is provided. The method includes steps of: (a) a training apparatus generating each feature vector corresponding to each of the segmented character images; (b) the training apparatus processing feature vectors ci+j of neighboring character images to thereby generate a support vector to be used for a recognition of a specific character image; (c) the training apparatus obtaining a merged vector by executing a computation with the support vector and a feature vector ci of the specific character image; and (d) the training apparatus (i) performing a classification of the specific character image as a letter included in a predetermined set of letters by referring to the merged vector; and (ii) adjusting the parameters by referring to a result of the classification.

The invention discloses an implicit discourse relation recognition method based on multi-granularity generated image enhancement representation, and provides a multi-granularity generated image and aneural network for enhancing argument vector representation by simulating an association strategy for the first time due to the problems of ambiguity, fuzziness and the like of texts. Corresponding images according to different granularities (sentence level and phrase level) of texts are introduced, which helps understand semantics of chapters. In order to better capture context information of a text image, text and image features are integrated according to the sequence information of the text .Important image-text information and interaction information are captured in an image-text vector sequence representation whole formed by splicing two arguments by utilizing a self-attention mechanism; therefore, argument vector representation is further enriched, feature vector representation usedfor recognizing the discourse relations is obtained, and finally the feature vector representation used for recognizing the discourse relations is input into the discourse relation recognition layerfor discourse relation recognition.

The invention discloses a electronic device that comprises: An image acquisition module, is used to acquire a text image to be input; a database module, which is a characteristic database at least with character sets and words to be identified; a central process module, according to the characteristic database that is stored in the database module, utilize a comparison identification algorithm to convert the text images that are acquired with the image acquisition module into the original text format. Accordingly, the invention also discloses a text input method for electronic devices. The invented electronic device and the relevant text input method can fulfill that, in a video-recording input mode, input the text into the electronic device and convert the text into a text format, so as to increase text input efficiency.

The embodiment of the invention provides an image processing method and device, electronic equipment and a readable storage medium, and aims to reduce the error rate of a character recognition result.The method comprises the following steps: carrying out target area detection on a to-be-processed image to obtain a target area containing a text image on the to-be-processed image; cutting the to-be-processed image according to the target area containing the text image to obtain a sub-image containing the text image; carrying out corner point prediction on the text image in the sub-images to obtain corner point position information of the text image; determining a correction parameter for the text image according to the angular point position information of the text image, and performing projection correction on the text image according to the correction parameter to obtain a target text image after projection correction; and performing character recognition on the target text image to obtain character information in the image.

The embodiment of the invention provides a text detection method and device, electronic equipment and a computer storage medium, and the method comprises the steps: carrying out the feature extractionand segmentation of a to-be-detected text image, and obtaining a text region threshold graph and a text region central point probability graph of the to-be-detected text image; obtaining a text areaframe binary image corresponding to the text area threshold image and a text area center point binary image corresponding to the text area center point probability image; and carrying out connected domain detection on the text area center point binary image, determining a clustering center of the text area, and further determining a text detection result in the to-be-detected text image accordingto the similarity between pixel points of a text area frame in the text area frame binary image and the clustering center. Through the method of the invention, the speed of text detection, especiallydense text detection, is increased.

The embodiment of the invention discloses a text detection method, electronic equipment and a computer readable medium, and the text detection method comprises the steps: carrying out the feature extraction and image segmentation of a to-be-detected text image, and at least obtaining a text region probability graph of the to-be-detected text image and the image features of the to-be-detected textimage; binarizing the text region probability graph to obtain a text region binary graph; obtaining at least one text connected domain according to the text region binary image and the image features;obtaining a text region approximate boundary of at least one text connected domain; and obtaining a text detection result of the to-be-detected text image according to the text region approximate boundary and a preset boundary threshold. According to the embodiment of the invention, the speed and efficiency of text detection, especially dense text detection, are improved.

The invention relates to a text image correction method and device, a computer device and a storage medium. The method comprises the steps of obtaining a text image needing to be corrected; inputtinga text image to be corrected into the edge detection model for deep learning to obtain a binary edge image; wherein the edge detection model is obtained by training a deep learning network through a plurality of text images with edge information; performing text fine positioning on the binary edge image to obtain a text bounding box and vertex coordinates of the text bounding box; calculating a perspective transformation matrix according to the vertex coordinates of the text bounding box; performing perspective transformation on the image in the text bounding box and the vertex coordinates ofthe text bounding box according to the perspective transformation matrix to obtain a corrected text image; and sending the corrected text image to the terminal for the terminal to consult and call. According to the method, the document edge extraction precision under different scenes is realized, the interference of noise edges of non-document regions is reduced, and the document coordinate position is refined.

The invention discloses a text recognition model training method, a text recognition method, device and equipment, and belongs to the technical field of image recognition, and the text recognition model training method comprises the steps: obtaining an image sample set, an image sample in the image sample set comprising a text image and a text label associated with the text image; performing sample expansion on the image sample set, and dividing the image sample set after sample expansion into a training set, a verification set and a test set; performing iterative training on a text recognition model according to the training set and the verification set, the text recognition model being constructed by replacing an original VGG network in a CRNN network model with an SE-ResNet network andsequentially cascading the SE-ResNet network with a BiLSTM network layer and an attention mechanism layer; and performing performance test on the text recognition model after iterative training according to the test set. According to the embodiment of the invention, the feature extraction capability of the text recognition model can be improved, and the feature vector decoding effect is improved,so that the text recognition accuracy is improved.

The invention relates to an American license plate recognition method and system based on image correction, and a text detection, image correction, text recognition and text classification module, andthe method comprises the following steps: carrying out the preprocessing of an image file of a data set, carrying out the data enhancement, and generating a training set and a test set; designing a text detection module, detecting text information in the image, realizing text and background segmentation in the image, and obtaining a text image only containing the text information; correcting thetext image by adopting an image correction module, and converting the originally distorted or inclined text image into a horizontal direction; recognizing the corrected text image to obtain letters, numbers and other information contained in the text image; and constructing a text classification module, screening out a license plate number, an Asian name and an annual inspection date from all textinformation, and completing license plate recognition. According to the invention, the problems of complex background pattern, fuzzy target text image deformation, complex text information and largecalculation amount when a neural network is used for off-line training during American license plate recognition are solved.

The invention discloses a multi-scale convolution kernel method based on a text-image generative adversarial network model. The method comprises the following steps that: S1: constructing the text-image generative adversarial network model; S2: utilizing a deep convolutional neural network to serve as the functions of a generator and a discriminator; S3: after a text is coded, combining with random noise, and inputting the combined text and random noise into the generator; S4: in the text-image generative adversarial network model, utilizing multi-scale convolution to carry out a convolution operation on an image; and S5: inputting a loss function obtained by the multi-scale convolution operation into the generator for subsequent training. By use of the text-image generative adversarial network model constructed by the method, a convolution way generated after the generator and the discriminator receive pictures is changed through the multi-scale convolution, an original operation thatonly one convolution kernel is used by aiming at a single-layer image channel is changed into a situation that a plurality of convolution kernels are simultaneously adopted, so that the whole networkcan learn more characteristics when the single-layer image channel is convoluted, and network training efficiency is improved.

The invention relates to a method for tracing text and strengthening multi-frame in a video. The texts in the video are mostly laminated in a complicated background, and if the texts are directly delivered into OCR (Optical Character Recognition) software for recognition, the recognition rate is low, therefore, the text strengthening operations are required to separate the texts from the background. The texts in the video mostly last for tens of frames, and even hundreds of frames, and in the adjacent frames, the colour of the texts are basically invariable but the background changes dynamically, accordingly, abundant complementary information between the multi-frames can be used for strengthening the texts. The method of the invention eliminates the influence of background edge pixel through the text stroke's characteristic of possession of edge pair, then uses Hausdorff distance measurement method for tracing the location of the texts at the adjacent frame, after obtaining a plurality of copies of the text images at the adjacent frame, makes use of minimum pixel searching method to remove the background in order to acquire binary text image including the clean background, greatly improving the rate of OCR software in identifying the video texts.

The embodiment of the invention discloses a text detection method, electronic equipment and a computer readable medium, and the text detection method comprises the steps: carrying out the feature extraction and segmentation of a to-be-detected text image, and obtaining a text region probability graph of the to-be-detected text image; determining a text region binary image of the to-be-detected text image according to the text region probability graph; extracting edge information of the text area binary image to obtain a text area edge image; performing connected domain detection on the text area edge graph, and obtaining a minimum enclosing rectangle of the text area according to a detection result; and obtaining a text detection result of the to-be-detected text image according to the minimum bounding rectangle. According to the method of the invention, the speed and efficiency of text detection, especially dense text detection, are improved.

The invention discloses a natural scene text recognition method based on a cross attention mechanism, and the method comprises the steps: data obtaining: downloading a sample picture in a natural scene, and synthesizing the picture into a training set through employing a public code; wherein stretching operation is conducted on the sizes of all the training sample pictures, the size of the processed sample pictures is 32 * 100, the height-width ratio is kept consistent with that of an original picture, and the insufficient parts are filled with black edges; label manufacturing: a supervised method is adopted to train an identification model, so that each row of text pictures has corresponding text information; training a network: inputting the prepared training picture data and labels intoa cross attention network for training, wherein the cross attention network is composed of a vertical attention network and a horizontal attention network; inputting test data into the trained network, and finally obtaining an identification result and predicting the confidence coefficient of each character. The method is high in recognition accuracy and strong in robustness, and has good recognition performance for texts with irregular shapes.