CN109033321B - Image and natural language feature extraction and keyword-based language indication image segmentation method - Google Patents
Image and natural language feature extraction and keyword-based language indication image segmentation method Download PDFInfo
- Publication number
- CN109033321B CN109033321B CN201810790480.4A CN201810790480A CN109033321B CN 109033321 B CN109033321 B CN 109033321B CN 201810790480 A CN201810790480 A CN 201810790480A CN 109033321 B CN109033321 B CN 109033321B
- Authority
- CN
- China
- Prior art keywords
- image
- word
- features
- language
- natural language
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Machine Translation (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method for extracting image and natural language features and segmenting a language indication image based on keywords, which is characterized in that on the basis of image feature extraction and natural language feature extraction, for an input image and an input natural language, the keywords are corresponding to the features f of an image area i according to the keywords contained in the natural languageiKeyword weighted sentence feature qiAnd corresponding keyword-based visual context features ciA total of three features are cascaded; and inputting the cascaded features into a multilayer perceptron for classification to obtain a segmentation result. Compared with the prior art, the method has the advantages that the characteristics of the image and the natural language are extracted, so that the method for segmenting the language indication image based on the key words can be conveniently realized; the language indication image segmentation method reduces the processing difficulty of long sentences, improves the accuracy of object positioning and identification, and further improves the segmentation precision of the language indication image.
Description
Technical Field
The invention relates to a method for extracting image and natural language features and segmenting a language indication image based on a keyword, which relates to the fields of image processing, computer vision, image segmentation and joint processing of language and images.
Background
With the advent of the big data era, mass data of different types flow in networks, and combining data of different types is a new requirement of the big data era. Among them, image processing has received a wide attention in combination with natural language. The language-specified image segmentation is to segment an object described in a natural language in an image, and is a key step in the joint processing of a language and an image.
The existing technology for solving the problem of segmenting the language indication image mainly utilizes a deep neural network to respectively extract natural language and image characteristics, and then combines the natural language and the image characteristics as new characteristics to segment the image. The method can be divided into two types: a sentence-based language indication image segmentation method and a word-based language indication image segmentation method. Extracting the characteristics of the whole sentence based on the language indication image segmentation method of the sentence, and combining the characteristics with the image characteristics; the word-based language indication image segmentation method extracts the features of each word and combines the features with the image features respectively. These methods have two major drawbacks:
1. neglecting the difference of importance between words, equally processing each word, resulting in difficult processing for long sentences;
2. the appearance, position and other contextual relationships between different regions inside the image are not considered, and these visual contextual relationships are often crucial to finding natural language described objects in the image.
Disclosure of Invention
The invention provides an image and natural language feature extraction method, which has the characteristic of being convenient for realizing a language indication image segmentation method based on key words.
The invention also provides a method for segmenting the language indication image based on the keywords, which has the characteristics of reducing the processing difficulty of long sentences and improving the accuracy of object positioning identification.
The invention provides an image and natural language feature extraction method, which comprises an image feature extraction method and a natural language feature extraction method; wherein the content of the first and second substances,
the image feature extraction method comprises the steps of extracting image features F from an input image by adopting a depth convolution neural network; the image feature is a two-dimensional feature map with each feature vector fiEncoding the characteristics of the corresponding area i in the image; indicating the position information of the object required by the image segmentation task according to the natural language characteristics;
the natural language feature extraction method comprises the steps of encoding each word into a one-hot feature vector for the input natural language, and then performing dimension reduction by word embedding; word press after dimension reductionInputting the sequence in the original sentence into a recurrent neural network in sequence; for the t-th word in the sentence, the recurrent neural network learns the feature q of the wordt(ii) a Features q of said wordtThe semantic information of the word t itself and the context information of the word t itself and the whole sentence are encoded; the feature vectors of the different words form a matrix Q representing the features of the whole sentence.
The method for indicating the position information of the object required by the image segmentation task according to the natural language characteristics comprises the steps of extracting the relative position coordinates of each image area, and cascading the relative position coordinates with the characteristics F to obtain the final visual characteristics V of each image area.
The method for segmenting the language indication image based on the keywords is realized based on the image and the natural language feature extraction method, and comprises the following steps,
for an input image and an input natural language, according to a keyword contained in the natural language, the keyword is associated with the feature f of the image area iiKeyword weighted sentence feature qiAnd corresponding keyword-based visual context features ciA total of three features are cascaded; inputting the cascaded features into a multilayer perceptron for classification to obtain a segmentation result;
the multilayer perceptron is composed of two layers of neural networks, wherein the first layer comprises a ReLU activation function, and the second layer comprises a sigmoid activation function;
wherein, the method for acquiring the image areas i corresponding to the keywords comprises the steps of training and extracting the keywords corresponding to each image area i, the process of training and extracting comprises,
extracting keywords by using a language attention model according to the obtained characteristics of each word; the language attention model is composed of two layers of neural networks, wherein the first layer comprises a tanh activation function, and the second layer has no activation function; for each image area i, firstly cascading each word t with the characteristics of the image area, and then inputting a language attention model to obtain an attention score; normalizing the attention scores, wherein the normalized score value is between 0 and 1, and the closer to 1, the more critical the word t is to the image area i; conversely, a closer to 0 indicates that the word t is less important for the image region i;
the characteristics of the modified sentences are scored by attention, so that the influence of the keywords in the sentences is improved, and the influence of the non-keywords is reduced; the normalized attention score is assigned to the corresponding word feature qtMultiplying, and weighting the word features; then adding the weighted features of all the words to generate a sentence feature q of the whole sentence for the image area ii；
Setting a keyword screening threshold, and if the normalized attention score is larger than the threshold, indicating that the word t is regarded as a keyword by the image area i;
for each word t, finding out all image areas which are regarded as keywords, and learning the context relationship of the areas; firstly, averaging the characteristics of the regions for integrating region information; then learning context features g based on the averaged features using a full connectivity layert ；
Learning the visual context characteristics g corresponding to each keywordtThen, integrating the visual context characteristics into the visual context characteristics corresponding to the whole sentence; for the image area i, the visual context characteristic g of each corresponding keywordtAdding to generate the visual context characteristic c corresponding to the whole sentencei。
The method further comprises normalizing the attention scores with softmax.
The method further comprises setting a keyword screening threshold value to 0.05.
Compared with the prior art, the method has the advantages that the characteristics of the image and the natural language are extracted, so that the method for segmenting the language indication image based on the key words can be conveniently realized; the language indication image segmentation method reduces the processing difficulty of long sentences, improves the accuracy of object positioning and identification, and further improves the segmentation precision of the language indication image.
Drawings
Fig. 1 is a schematic diagram of an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Any feature disclosed in this specification (including any accompanying drawings) may be replaced by alternative features serving equivalent or similar purposes, unless expressly stated otherwise. That is, unless expressly stated otherwise, each feature is only an example of a generic series of equivalent or similar features.
An image and natural language feature extraction method comprises an image feature extraction method and a natural language feature extraction method; wherein the content of the first and second substances,
the image feature extraction method comprises the steps of extracting an image feature F from an input image by adopting a deep Convolutional Neural Network (CNN); the image feature is a two-dimensional feature map with each feature vector fiEncoding the characteristics of the corresponding area i in the image; indicating the position information of the object required by the image segmentation task according to the natural language characteristics;
the natural language feature extraction method comprises the steps of coding each word into one-hot feature vectors and then performing dimension reduction by word embedding (word embedding) for the input natural language; the words after dimensionality reduction are sequentially input into a Recurrent Neural Network (RNN) according to the sequence in the original sentence; for the t-th word in the sentence, the recurrent neural network learns the feature q of the wordt(ii) a Features q of said wordtThe semantic information of the word t itself and the context information of the word t itself and the whole sentence are encoded; the feature vectors of the different words form a matrix Q representing the features of the whole sentence.
Based on the image feature extraction method and the natural language feature extraction method, the language indication image segmentation method based on the keywords is convenient to realize.
As an embodiment of the present invention, the method for indicating the position information of the object required by the image segmentation task according to the natural language features includes extracting the relative position coordinates of each image region, and cascading the relative position coordinates with the feature F to obtain the final visual feature V of each image region.
As shown in FIG. 1, a method for segmenting language indication image based on keywords is realized based on the image and natural language feature extraction method, and comprises,
for an input image and an input natural language, according to a keyword contained in the natural language, the keyword is associated with the feature f of the image area iiKeyword weighted sentence feature qiAnd corresponding keyword-based visual context features ciA total of three features are cascaded; inputting the cascaded features into a multilayer perceptron (MLP) for classification to obtain a segmentation result;
the multilayer perceptron is composed of two layers of neural networks, wherein the first layer comprises a ReLU activation function, and the second layer comprises a sigmoid activation function;
wherein, the method for acquiring the image areas i corresponding to the keywords comprises the steps of training and extracting the keywords corresponding to each image area i, the process of training and extracting comprises,
extracting keywords by using a language attention model according to the obtained characteristics of each word; the language attention model is composed of two layers of neural networks, wherein the first layer comprises a tanh activation function, and the second layer has no activation function; for each image area i, firstly cascading each word t with the characteristics of the image area, and then inputting a language attention model to obtain an attention score; normalizing the attention scores, wherein the normalized score value is between 0 and 1, and the closer to 1, the more critical the word t is to the image area i; conversely, a closer to 0 indicates that the word t is less important for the image region i;
the characteristics of the modified sentences are scored by attention, so that the influence of the keywords in the sentences is improved, and the influence of the non-keywords is reduced; the normalized attention score is assigned to the corresponding word feature qtMultiplying, and weighting the word features; then adding the weighted features of all the words to generate a sentence feature q of the whole sentence for the image area ii；
Setting a keyword screening threshold, and if the normalized attention score is larger than the threshold, indicating that the word t is regarded as a keyword by the image area i;
for each word t, finding out all image areas which are regarded as keywords, and learning the context relationship of the areas; firstly, averaging the characteristics of the regions for integrating region information; then learning context features g based on the averaged features using a full connectivity layert ；
Learning the visual context characteristics g corresponding to each keywordtThen, integrating the visual context characteristics into the visual context characteristics corresponding to the whole sentence; for the image area i, the visual context characteristic g of each corresponding keywordtAdding to generate the visual context characteristic c corresponding to the whole sentencei。
In the prior art, on one hand, each word in a sentence is processed equally, so that a long sentence is difficult to process; on the other hand, the context of appearance, location, etc. between different regions within the image is not taken into account, which is crucial for locating and identifying objects described in natural language in the image. The invention provides a keyword-based language indication image segmentation algorithm, and the keywords in the natural language are extracted, so that the processing difficulty of long sentences is reduced. And the visual context relation based on the key words is learned, so that the accuracy of object positioning and identification is improved, and the segmentation precision of the language indication image is further improved.
As an embodiment of the present invention, the method further includes performing normalization processing on the attention score using softmax.
As an embodiment of the present invention, the method further includes setting a keyword screening threshold to 0.05.
The following is a further detailed description of an embodiment.
And determining a database, and determining a language indication image segmentation database, such as a Google reference database.
And data preprocessing, namely preprocessing the database, extracting an original image, a natural language and a segmentation result. The relative position coordinates of each point need to be extracted from the original image. Natural language requires that each word in a sentence be converted into a one-hot vector.
And (5) building a deep network model. The Convolutional Neural Network (CNN) is derived from deep lab101, which outputs 60 × 60 image regions, and the feature fi of each region is set to 1000 dimensions. The long-time and short-time memory unit (LSTM) is selected as the Recurrent Neural Network (RNN), the maximum word number of each sentence is set as 20, and the feature qt of each word is set as 1000 dimensions.
A keyword threshold is determined. The keyword threshold Thr is set to 0.05.
Model initialization, the Convolutional Neural Network (CNN) initializes the parameters with the pre-trained model on ImageNet. The rest of the model is initialized randomly.
Learning rate and gradient descent strategy are set, the learning rate of a Convolutional Neural Network (CNN), a language attention model, a keyword-based visual context learning model and a multilayer perceptron (MLP) is set to be 0.0001, and the learning rate of a Recurrent Neural Network (RNN) is set to be 0.001. The optimization mode adopts an ADAM gradient descent strategy.
And training the model, wherein after the model is built and initialized, the learning and gradient descent strategy is determined, and then training is carried out. And sequentially inputting the data of the training set in the database into the model, and training 5 epochs.
And testing the model, inputting the images and sentences of the test set in the database after the model training is finished, and obtaining the language indication image segmentation result.
Claims (4)
1. A method for segmenting language indication image based on key words is realized based on image and natural language feature extraction method, and comprises,
for an input image and an input natural language, according to a keyword contained in the natural language, the keyword is associated with the feature f of the image area iiKeyword weighted sentence feature qiAnd corresponding keyword-based visual context features ciA total of three features are cascaded; inputting the cascaded features into a multilayer perceptron for classification to obtain a segmentation result;
the multilayer perceptron is composed of two layers of neural networks, wherein the first layer comprises a ReLU activation function, and the second layer comprises a sigmoid activation function;
wherein, the method for acquiring the image areas i corresponding to the keywords comprises the steps of training and extracting the keywords corresponding to each image area i, the process of training and extracting comprises,
extracting keywords by using a language attention model according to the obtained characteristics of each word; the language attention model is composed of two layers of neural networks, wherein the first layer comprises a tanh activation function, and the second layer has no activation function; for each image area i, firstly cascading each word t with the characteristics of the image area, and then inputting a language attention model to obtain an attention score; normalizing the attention scores, wherein the normalized score value is between 0 and 1, and the closer to 1, the more critical the word t is to the image area i; conversely, a closer to 0 indicates that the word t is less important for the image region i;
the characteristics of the modified sentences are scored by attention, so that the influence of the keywords in the sentences is improved, and the influence of the non-keywords is reduced; the normalized attention score is assigned to the corresponding word feature qtMultiplying, and weighting the word features; then adding the weighted features of all the words to generate a sentence feature q of the whole sentence for the image area ii；
Setting a keyword screening threshold, and if the normalized attention score is larger than the threshold, indicating that the word t is regarded as a keyword by the image area i;
for each word t, finding out all image areas which are regarded as keywords, and learning the context relationship of the areas; firstly, averaging the characteristics of the regions for integrating region information; then learning context features g based on the averaged features using a full connectivity layert；
Learning the visual context characteristics g corresponding to each keywordtThen, integrating the visual context characteristics into the visual context characteristics corresponding to the whole sentence; for the image area i, the visual context characteristic g of each corresponding keywordtAdding to generate visual context character corresponding to whole sentenceSymbol ci；
The method comprises an image feature extraction method and a natural language feature extraction method; wherein the content of the first and second substances,
the image feature extraction method comprises the steps of extracting image features F from an input image by adopting a depth convolution neural network; the image feature is a two-dimensional feature map with each feature vector fiEncoding the characteristics of the corresponding area i in the image; indicating the position information of the object required by the image segmentation task according to the natural language characteristics;
the natural language feature extraction method comprises the steps of encoding each word into a one-hot feature vector for the input natural language, and then performing dimension reduction by word embedding; the words after dimensionality reduction are sequentially input into a recurrent neural network according to the sequence in the original sentence; for the t-th word in the sentence, the recurrent neural network learns the feature q of the wordt(ii) a Features q of said wordtThe semantic information of the word t itself and the context information of the word t itself and the whole sentence are encoded; the feature vectors of the different words form a matrix Q representing the features of the whole sentence.
2. The method for segmenting the language-indicating image according to claim 1, wherein the method for indicating the position information of the object required by the image segmentation task according to the natural language features comprises the steps of extracting the relative position coordinates of each image region, and cascading the relative position coordinates with the features F to obtain the final visual features V of each image region.
3. The method of language-specific image segmentation as set forth in claim 1, further comprising normalizing the attention scores with softmax.
4. The method of segmenting a language-specific image according to claim 1 or 3, further comprising setting a keyword filtering threshold of 0.05.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810790480.4A CN109033321B (en) | 2018-07-18 | 2018-07-18 | Image and natural language feature extraction and keyword-based language indication image segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810790480.4A CN109033321B (en) | 2018-07-18 | 2018-07-18 | Image and natural language feature extraction and keyword-based language indication image segmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109033321A CN109033321A (en) | 2018-12-18 |
| CN109033321B true CN109033321B (en) | 2021-12-17 |
Family
ID=64643921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810790480.4A Active CN109033321B (en) | 2018-07-18 | 2018-07-18 | Image and natural language feature extraction and keyword-based language indication image segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109033321B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711463B (en) * | 2018-12-25 | 2023-04-07 | 广东顺德西安交通大学研究院 | Attention-based important object detection method |
| CN111598155A (en) * | 2020-05-13 | 2020-08-28 | 北京工业大学 | Fine-grained image weak supervision target positioning method based on deep learning |
| CN112037239B (en) * | 2020-08-28 | 2022-09-13 | 大连理工大学 | Text guidance image segmentation method based on multi-level explicit relation selection |
| CN114299348A (en) * | 2022-02-21 | 2022-04-08 | 山东力聚机器人科技股份有限公司 | Image classification method and device based on restoration self-supervision task |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559193A (en) * | 2013-09-10 | 2014-02-05 | 浙江大学 | Topic modeling method based on selected cell |
| CN106599198A (en) * | 2016-12-14 | 2017-04-26 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Image description method for multi-stage connection recurrent neural network |
| CN106778835A (en) * | 2016-11-29 | 2017-05-31 | 武汉大学 | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic |
| CN107391709A (en) * | 2017-07-28 | 2017-11-24 | 深圳市唯特视科技有限公司 | A kind of method that image captions generation is carried out based on new attention model |
| CN107688821A (en) * | 2017-07-11 | 2018-02-13 | 西安电子科技大学 | View-based access control model conspicuousness and across the modality images natural language description methods of semantic attribute |
| US9939272B1 (en) * | 2017-01-06 | 2018-04-10 | TCL Research America Inc. | Method and system for building personalized knowledge base of semantic image segmentation via a selective random field approach |
| CN107909115A (en) * | 2017-12-04 | 2018-04-13 | 上海师范大学 | A kind of image Chinese subtitle generation method |
| CN108009154A (en) * | 2017-12-20 | 2018-05-08 | 哈尔滨理工大学 | A kind of image Chinese description method based on deep learning model |
| CN108228686A (en) * | 2017-06-15 | 2018-06-29 | 北京市商汤科技开发有限公司 | It is used to implement the matched method, apparatus of picture and text and electronic equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106227851B (en) * | 2016-07-29 | 2019-10-01 | 汤一平 | The image search method of depth of seam division search based on depth convolutional neural networks |
| CN107608943B (en) * | 2017-09-08 | 2020-07-28 | 中国石油大学(华东) | Image subtitle generating method and system fusing visual attention and semantic attention |
-
2018
- 2018-07-18 CN CN201810790480.4A patent/CN109033321B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559193A (en) * | 2013-09-10 | 2014-02-05 | 浙江大学 | Topic modeling method based on selected cell |
| CN106778835A (en) * | 2016-11-29 | 2017-05-31 | 武汉大学 | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic |
| CN106599198A (en) * | 2016-12-14 | 2017-04-26 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Image description method for multi-stage connection recurrent neural network |
| US9939272B1 (en) * | 2017-01-06 | 2018-04-10 | TCL Research America Inc. | Method and system for building personalized knowledge base of semantic image segmentation via a selective random field approach |
| CN108228686A (en) * | 2017-06-15 | 2018-06-29 | 北京市商汤科技开发有限公司 | It is used to implement the matched method, apparatus of picture and text and electronic equipment |
| CN107688821A (en) * | 2017-07-11 | 2018-02-13 | 西安电子科技大学 | View-based access control model conspicuousness and across the modality images natural language description methods of semantic attribute |
| CN107391709A (en) * | 2017-07-28 | 2017-11-24 | 深圳市唯特视科技有限公司 | A kind of method that image captions generation is carried out based on new attention model |
| CN107909115A (en) * | 2017-12-04 | 2018-04-13 | 上海师范大学 | A kind of image Chinese subtitle generation method |
| CN108009154A (en) * | 2017-12-20 | 2018-05-08 | 哈尔滨理工大学 | A kind of image Chinese description method based on deep learning model |
Non-Patent Citations (3)
| Title |
|---|
| Attention to Scale: Scale-aware Semantic Image Segmentation;Liang-Chieh Chen et al.;《2016 IEEE Conference on Computer Vision and Pattern Recognition》;20161212;3640-3649 * |
| Segmentation from Natural Language;Ronghang Hu et al.;《European Conference on Computer Vision》;20160917;第9905卷;108-124 * |
| 图像检索中语义映射方法综述;李志欣 等;《计算机辅助设计与图形学学报》;20080815;第20卷(第8期);1085-1096 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109033321A (en) | 2018-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20200285896A1 (en) | Method for person re-identification based on deep model with multi-loss fusion training strategy | |
| CN107688821B (en) | Cross-modal image natural language description method based on visual saliency and semantic attributes | |
| CN107506761B (en) | Brain image segmentation method and system based on significance learning convolutional neural network | |
| CN109033321B (en) | Image and natural language feature extraction and keyword-based language indication image segmentation method | |
| WO2021135193A1 (en) | Visual object guidance-based social media short text named entity identification method | |
| CN103984943B (en) | A kind of scene text recognition methods based on Bayesian probability frame | |
| CN110490081B (en) | Remote sensing object interpretation method based on focusing weight matrix and variable-scale semantic segmentation neural network | |
| CN109446333A (en) | A kind of method that realizing Chinese Text Categorization and relevant device | |
| CN109190521B (en) | Construction method and application of face recognition model based on knowledge purification | |
| CN109684928B (en) | Chinese document identification method based on internet retrieval | |
| CN111325237B (en) | Image recognition method based on attention interaction mechanism | |
| CN112560710B (en) | Method for constructing finger vein recognition system and finger vein recognition system | |
| WO2019167784A1 (en) | Position specifying device, position specifying method, and computer program | |
| CN111862260B (en) | Cross-domain dual generation type countermeasure network-based bias eliminating method and device | |
| Alon et al. | Deep-hand: a deep inference vision approach of recognizing a hand sign language using american alphabet | |
| CN109145704A (en) | A kind of human face portrait recognition methods based on face character | |
| CN113076905B (en) | Emotion recognition method based on context interaction relation | |
| Setyono et al. | Recognizing word gesture in sign system for Indonesian language (SIBI) Sentences using DeepCNN and BiLSTM | |
| Drishya et al. | Cyberbully image and text detection using convolutional neural networks | |
| CN109657710B (en) | Data screening method and device, server and storage medium | |
| Jindal et al. | Sign Language Detection using Convolutional Neural Network (CNN) | |
| Revelli et al. | Automate extraction of braille text to speech from an image | |
| CN115497105A (en) | Multi-modal hate cause detection method based on multi-task learning network | |
| CN111160219B (en) | Object integrity evaluation method and device, electronic equipment and storage medium | |
| Paul et al. | An Adam based CNN and LSTM approach for sign language recognition in real time for deaf people |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |