60 results about "Text enhancement" patented technology

A text enhancement method and apparatus for the presentation of text for improved human reading. The method includes extracting text specific attributes from machine readable text and varying the text presentation in accordance with the attributes. The preferred embodiment of the method: extracts parts of speech and punctuation from a sentence, applies folding rules which use the parts of speech to determine folding points, and presents text segments each on a new line and having a determined horizontal displacement based on the text specific attributes. One method displays text over bent curves having a shape based on text content. Another method includes displaying relative text position within a hierarchy using alternating vertically and horizontally tiled planes. Another method supports reading text segments across opposed pages without waiting for paging. Yet another method displays text to allow reading from bottom to top as though from front to back. Still another method displays words in colors reflecting the relationships between the words and the larger text segments of which they are apart.

The invention discloses a text enhancement semantic classification method and system based on a convolutional neural network, and belongs to the technical field of natural language processing, and themethod comprises the following steps: S1, collecting a training sample; s2, performing pretreatment; S3, segmenting words S4, constructing a word segmentation matrix; S5, enhancing data S6, trainingby using model. A new text word vector matrix with the same label can be generated so that a small amount of label data in an original data set is enhanced to a great extent, the sample capacity is expanded, and the effects of improving the robustness of a subsequent model, improving the accuracy, the accuracy rate and the recall rate and the like are achieved; a model is trained through the improved convolutional neural network, and texts under government affair public opinion Chinese text labels can be effectively classified and judged; the invention is suitable for solving the semantic category classification problem of the Chinese text, and is also suitable for solving other classification problems such as sentiment dichotomy.

The invention discloses a video text extraction method, comprising: dividing a video sequence into a single shot; detecting and locating the specific position of a candidate text in a single video frame sequence; Track the text to obtain the text area sequence of the same text object in consecutive multiple frames; use the text sequence obtained by text tracking to enhance the text, suppress the background, and then perform binarization to obtain a binarized text image; The final text image is subjected to text recognition to obtain the character sequence information of the text. By adopting the technical scheme provided by the invention, the text in the video can be better detected and separated from the complex polygonal background, the system efficiency can be improved, and the quality of the text can be improved, thereby helping to improve the text recognition rate.

The invention belongs to the technical field of artificial intelligence and provides a text classification model training method and device, equipment and a storage medium. The method comprises the steps that: a training sample set is acquired, wherein the training sample set comprises N labeled training samples and M unlabeled training samples, wherein each labeled training sample comprises textinformation and a category label of the text information, and each unlabeled training sample comprises text information, M and N are integers greater than 1; alternate iterative training is performedon an initial text classification model and an initial text enhancement model according to the training sample set and M enhanced training samples to obtain a target text classification model, whereinin the ith alternate iteration training process, the M enhanced training samples are generated by performing text enhancement processing on the M unlabeled training samples according to a text enhancement model obtained through the i-1th alternate iteration, wherein i is an integer greater than 1. According to the text classification model training method provided by the embodiment of the invention, the performance of the finally obtained text classification model is improved.

The embodiment of the invention discloses a text enhancement method, a text enhancement device, a text extraction method and a text extraction device. The text enhancement method comprises the following steps: acquiring an original image including a line of text; according to the direct difference degree and indirect difference degree between any original pixel in the original image and each neighborhood pixel in a neighborhood set of the original image, carrying out stroke two-dimensional filtering on the original brightness value or/and color value of each original pixel so as to obtain an updated brightness value or/and color value of the filtered original image, wherein the range of the neighborhood set is a square with a side length of w and the original pixel as a center, and w is less than the height of the original image; and respectively replacing corresponding original brightness value or/and color value with the updated brightness value or/and color value after filtering so as to generate a text enhancement image corresponding to the original image. Through the embodiment of the invention, texts in original images can be enhanced, so that the subsequent text extraction on the text enhancement image is more precise and accurate.

A method for enhancing a textual image for undergoing optical character recognition begins by receiving an image that includes native lines of text. A background line profile is determined which represents an average background intensity along the native lines in the image. Likewise, a foreground line profile is determined which represents an average foreground background intensity along the native lines in the image. The pixels in the image are assigned to either a background or foreground portion of the image based at least in part on the background line profile and the foreground line profile. The intensity of the pixels designated to the background portion of the image is adjusted to a maximum brightness so as to represent a portion of the image that does not include text.

The invention provides a text enhancement method, a text classification method and a related device. The method comprises the steps: acquiring a statement content in a text corpus, performing word segmentation on the statement content, and obtaining words obtained after word segmentation; screening out similar words of which the similarity with the segmented words exceeds a threshold value from the concept tree, and randomly replacing the statement content by using the similar words to obtain a plurality of statements; training a generative adversarial network by using a plurality of statements to obtain a generative adversarial network model; generating an expansion statement sample by using a generative adversarial network model; and combining the expanded statement sample with the statement content in the text corpus to obtain an enhanced text data set. In the implementation process, the expanded statement sample is generated by using the generative adversarial network model obtained by training, and the generative adversarial network model learns the newly added change rule between the similar words with the similarity exceeding the threshold value in the concept tree, so a specific category sample can be better generated.

The embodiment of the invention discloses a text enhancement method and device, a computer device, and a readable storage medium so that a display effect of a text in an image can be improved and picture distortion of the image can be reduced. The method comprises: extracting a text region of a current frame of image; determining first pixel information of a first pixel point of the text region; acquiring second pixel information of a second pixel point in a preset range of the first pixel point; and adjusting the first pixel information based on the second pixel information.

The invention provides a speech recognition text enhancement system fused with multi-modal semantic invariance. The system comprises an acoustic feature extraction module, an acoustic downsampling module, an encoder and a decoder fused with multi-modal semantic invariance; A method comprises the following steps: performing framing processing on the voice data by the acoustic feature extraction module, dividing the voice data into short-time audio frames with fixed lengths, extracting acoustic features from the short-time audio frames, and inputting the acoustic features into the acoustic downsampling module for downsampling to obtain acoustic representation; inputting the voice data into an existing voice recognition module to obtain input text data, and inputting the input text data into an encoder to obtain input text encoding representation; and inputting the acoustic representation and the input text coding representation into a decoder for fusion, and carrying out similarity constraint on the representation of the acoustic mode and the representation of the text mode to obtain a decoding representation. According to the method, by fusing the cross-modal semantic invariance constraint loss, the dependence of the model on data is reduced, the performance of the model is improved, and the method is suitable for Chinese-English mixed speech recognition.

The present invention provides a text-picture separating method and a text enhancing method. The text-picture separating method is used for separating a picture part and a text part of an image, and comprises a rough classification step, a progressive classification step and a text-on-picture detection step. The text enhancing method is used for enhancing the text part in the image comprising the picture part and the text part, and comprises the following steps: a. a text-picture separation step for separating the picture part and the text part, comprising that the picture is cut into a plurality of lines and each line is cut into a plurality of segments which are classified into picture segments and text segments; b. a text enhancing step comprising that the text segments are sharpened to enhance the contrast grade of the text segments. The text-picture separating method and the text enhancing method carry out brightness adjustment on the obtained text part so as to enhance the image of the text part.

The invention discloses a text information feature extraction and recognition method based on a Gabor filter to solve the problem that an existing technology for extracting text information in videos or images is low in efficiency. The method includes the following steps that firstly, the Gabor filter is designed; secondly, a DBN classification network is designed and trained; thirdly, denoising, void region filling and isolated point removing are conducted in a localized image with the morphological method to make the localized text image more precise, the denoised text localization binary image is mapped to an original video frame image, and an accurate text localization region is obtained; fourthly, text enhancement, binarization processing, normalization and feature extraction are conducted on the accurate text localization region obtained after localization and processing; fifthly, the text processed in the fourth step is recognized with the OCR technology. By means of the method, text information in videos or images can be extracted more precisely.

The embodiment of the invention is applied to the field of artificial intelligence, and discloses a text classification model training method and device, equipment and a readable medium, and the method comprises the steps: obtaining a labeled sample set and a non-labeled sample set of a classification model; performing text enhancement processing on the second text data in the unmarked sample set to obtain enhanced text data; inputting the labeled sample set, the unlabeled sample set and the enhanced text data into a classification model; training the classification model according to a first loss function determined by the predicted first probability distribution of each piece of first text data in the labeled sample set, the predicted second probability distribution of each piece of second text data and the predicted third probability distribution of the enhanced text data, the labeled sample set and the unlabeled sample set; and when the first loss function meets a training ending condition, determining a target text classification model. By adopting the embodiment of the invention, the model training efficiency can be improved, and the service iteration speed can be improved. The invention relates to the blockchain technology. The data can be stored in a blockchain.

The present invention is to provide a text enhancement methodology of gray-scale levels of a scanned image so as to enhance text legibility. The text enhancement methodology includes the following steps of: utilizing a window block pattern as a fumdamental processing unit to read the scanned image, wherein the window block pattern comprises at least two pixels in a predetermined neighborhood; analyzing the scanned image in the window block pattern to find right, left, upper, and bottom edges; labeling the image in the window block pattern as two groups of text pixels and non-text pixels based on the window block pattern and the right, left, upper, and bottom edges; and enhancing the text pixels/non-text pixels of scanned image in the window block pattern based on the window block pattern and the right, left, upper, and bottom edges.

The invention relates to a text enhancement method and system based on word interpretation, and the method comprises the steps of obtaining a to-be-tested text, and obtaining an interpretation sentence of a target word in the to-be-tested text; then preprocessing a to-be-tested text, for a text classification task taking the target word as a label, setting an explaining sentence of the target word as a label, and for a text classification task not taking the target word as the label, adding the explaining sentence of the target word into the text; and finally, training the natural language classification model through the preprocessed text, so that the problems that the model training effect is poor and the accuracy rate is low when the text is classified and words with insufficient related corpora such as new words exist in the text are solved, and the accuracy rate of the model is improved.

The invention provides a remote supervision relation extraction method and device based on consistency text enhancement, and the method comprises the steps: dividing a plurality of sentence instances according to entity pairs and relation labels, and obtaining a plurality of sentence packages; adopting different text enhancement methods for each sentence instance in each sentence packet, and obtaining a strong enhancement sample and a weak enhancement sample corresponding to each sentence instance in each sentence packet; determining a noise sample, and training the relation prediction model through the unrelated sentence instance and the strong enhancement sample and the weak enhancement sample of the noise sample to obtain a trained relation prediction model; and predicting a to-be-predicted sentence package by using the trained relationship prediction model to obtain a relationship label corresponding to the to-be-predicted sentence package. According to the method, the scale of a data set can be increased through consistent text enhancement, the generalization learning ability of the model can be enhanced, and the NA category and noise sample constraint model can learn more supervision information.

The present invention provides a text and graphic separation method and a text enhancement method. The text and graphic separation method is used for separating texts and graphics of an image and comprises coarse classification and advanced classification. The method of the present invention also adjusts the luminance of the text to enhance the text image according to the separation result.

The invention provides a text classification method, terminal equipment and a computer readable storage medium, and the method comprises the steps: adding a full connection layer on the basis of a first model, and carrying out the training of the parameters of the full connection layer based on a supervised data set, and obtaining a fine adjustment pre-training model; obtaining an enhanced data set with doubled data volume based on the supervised data set through a text enhancement algorithm, and classifying the enhanced data set through a fine tuning pre-training model to obtain first data distribution of the enhanced data set; training a second model by utilizing a supervised data set, classifying the enhanced data set through the second model to obtain second data distribution of the enhanced data set, wherein the parameter quantity of the first model is more than ten times that of the second model; and further training the second model by maximizing the similarity between the firstdata distribution and the second data distribution, and storing all parameters for classifying the to-be-classified text. And the accuracy is improved on the basis of ensuring that the rate of the second model is unchanged.

The invention relates to the field of artificial intelligence, and provides a user tag extraction method and device, equipment and a medium, which can recall all possible text expressions through text enhancement, and then screen the text obtained after enhancement through sensitive word recall so as to ensure that keywords are accurately positioned as many as possible for subsequent model training. The model training effect is improved. A user tag extraction model is obtained by combining sensitive word recall and deep learning training; meanwhile, the high adaptability of sensitive word recognition in a public opinion scene and the high accuracy of a deep learning algorithm for text sentiment tone judgment are fused. A large number of recalled public opinion related texts are recognized through sensitive words, and the emotional positivity and negativity of the recalled text are judged through deep learning, so that the accuracy of tag identification can be effectively improved. In addition, the invention also relates to a block chain technology, and the user tag extraction model can be stored in a block chain node.

The invention provides a knowledge distillation-based confidential text recognition model training method, system and device. The method comprises the following steps: preparing an untagged corpus A in a confidentiality field; constructing a text label hierarchical tree according to confidential business data, labeling to obtain a labeled data set B, and preparing a non-labeled data set C; performing text enhancement on the texts in the label-free data set C; performing knowledge distillation through the label-free corpus A, so that the IDCNN model learns the semantic feature extraction capability from the Bert model; constructing a label path classification model Bert-clf based on a Bert model, and performing supervised classification training on the Bert model through the labeled data set B to obtain the label path classification model Bert-clf; building a label path classification model Idcnn-clf based on the IDCNN model; performing knowledge distillation on the label path classification model Idcnn-clf through the label data set B and the label-free data set C; and storing the label path classification model Idcnn-clf. According to the invention, the prediction speed and the classification accuracy of the classified text recognition model can be effectively improved.

The invention provides a case document cause classification method based on legal instrument graph convolutional network text enhancement, which comprises the following steps: 1, data preprocessing: carrying out case extraction, text word segmentation and noise reduction and legal instrument co-occurrence relationship graph construction on data; 2, legal word embedding pre-training: carrying out a pre-training task on the legal domain corpus set to obtain legal domain word embedding; 3, model construction: establishing a case document cause classification model based on law bar chart convolution text enhancement; 4, model training: performing gradient descent update training on the model constructed in the step 3 by training set data to obtain model parameters; 5, case cause prediction: performing, by the classification model trained in the step 4, case cause classification on the case situation description text to be classified. According to the case document cause classification method based on the legal provision convolutional network text and the medium provided by the invention, enhanced legal provision text representation has stronger cause correlation.

Auxiliary or enhanced features associated with broadcast television programs are activated using carrier-based active text enhancement (CATE) signals embedded within timed text (TT) associated with the broadcast program. The active text enhancements can be interpreted by the viewer's set top box (STB) or other receiver to activate software applications, video clips, imagery, uniform resource locators (URLs), interactive interface features or the like on either or both of primary or secondary displays. Timed text enhancements can flexibly reference different types of content to provide richer and more powerful viewer experiences for the viewer.

The invention discloses a task type dialogue text enhancement system, and relates to the technical field of text processing. The task type dialogue text enhancement system comprises a synonym replacement module, a synonym replacement module, an entity replacement module, a noise injection module and a text mixing module; the synonym replacement module is used for replacing corresponding words withsynonyms of each word in a text to generate first training data; the synonym replacement module is used for replacing the corresponding words with synonyms of the words in the text to generate secondtraining data; the entity replacement module is used for replacing corresponding entities with entities of the same type as the entities in the text to generate third training data; the noise injection module is used for emptying, exchanging, adding and deleting words of the text to generate fourth training data; and the text mixing module is used for mixing the first training data, the second training data, the third training data and the fourth training data to obtain fifth training data, so that the problem of insufficient training data is relieved, and the performance of the task type dialogue system is further optimized.