Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993497
Zakia Kezzoula, Djamel Gaceb, Nadjat Gritli
This paper presents a novel super-resolution approach of document images. It is based on transfer deep learning of an ESRGAN model. This model, which showed good robustness on natural images, has been adapted to document images by using better levels of fine-tuning and a post-processing to enhance contrast. The experiments were carried out on our document image dataset that we built from document images presenting different challenges. Documents of different categories with different complexity levels and degradation kinds. The results obtained are better compared to ten existing approaches, which we have developed and tested on the same dataset with the same evaluation protocol.
{"title":"Super-resolution of document images using transfer deep learning of an ESRGAN model","authors":"Zakia Kezzoula, Djamel Gaceb, Nadjat Gritli","doi":"10.1109/ISIA55826.2022.9993497","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993497","url":null,"abstract":"This paper presents a novel super-resolution approach of document images. It is based on transfer deep learning of an ESRGAN model. This model, which showed good robustness on natural images, has been adapted to document images by using better levels of fine-tuning and a post-processing to enhance contrast. The experiments were carried out on our document image dataset that we built from document images presenting different challenges. Documents of different categories with different complexity levels and degradation kinds. The results obtained are better compared to ten existing approaches, which we have developed and tested on the same dataset with the same evaluation protocol.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"515 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116213442","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993577
G. Chekhmane, R. Benali
Electroencephalography (EEG) is an important tool for diagnosis of brain disorders such as epilepsy, it can measure the electrical activity of neurons and the recorded signal includes different characteristics in order to detect epileptic seizures. In this study, the analysis of the EEG signals was based on the Discrete Wavelet Transform (DWT) and some statistical features were extracted from the sub-bands to be as inputs in the Machine Learning (ML), by using two different classifiers: the Support Vector Machine (SVM) and Multilayer Perceptron Neural Network (MLPNN) for the automatic detection of this disease. Then, the performance of the classification process of both methods was presented and the results obtained by SVM and MLPNN are 99.5% and 100% of accuracy, respectively. Finally, our study shows that the two methods perform better in the detection of epilepsy and that the MLPNN achieved a higher accuracy.
{"title":"EEG signals analysis using SVM and MLPNN classifiers for epilepsy detection","authors":"G. Chekhmane, R. Benali","doi":"10.1109/ISIA55826.2022.9993577","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993577","url":null,"abstract":"Electroencephalography (EEG) is an important tool for diagnosis of brain disorders such as epilepsy, it can measure the electrical activity of neurons and the recorded signal includes different characteristics in order to detect epileptic seizures. In this study, the analysis of the EEG signals was based on the Discrete Wavelet Transform (DWT) and some statistical features were extracted from the sub-bands to be as inputs in the Machine Learning (ML), by using two different classifiers: the Support Vector Machine (SVM) and Multilayer Perceptron Neural Network (MLPNN) for the automatic detection of this disease. Then, the performance of the classification process of both methods was presented and the results obtained by SVM and MLPNN are 99.5% and 100% of accuracy, respectively. Finally, our study shows that the two methods perform better in the detection of epilepsy and that the MLPNN achieved a higher accuracy.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130075567","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993553
R. Bouguesri, Khadidja Habelhames, H. Aliane, A. A. Aliane
Sarcasm is one of the main challenges of sentiment analysis systems. This paper mainly focuses on the recognition of Arabic sarcasm on Twitter. Recognizing sarcasm in tweets is essential for understanding users' opinions on various topics and events. There are only a few attempts regarding saracsm detection in Arabic due to the challenges and complexity of the Arabic language. We propose in this paper a comparison between traditional neural network-based models and pre-trained transformers. The experimental results show that transformers are a promising approach for the task of Arabic sarcasm detection.
{"title":"Sarcasm Detection in Arabic Tweets: A comparison Between deep learning and Pre trained Transformers-based Models","authors":"R. Bouguesri, Khadidja Habelhames, H. Aliane, A. A. Aliane","doi":"10.1109/ISIA55826.2022.9993553","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993553","url":null,"abstract":"Sarcasm is one of the main challenges of sentiment analysis systems. This paper mainly focuses on the recognition of Arabic sarcasm on Twitter. Recognizing sarcasm in tweets is essential for understanding users' opinions on various topics and events. There are only a few attempts regarding saracsm detection in Arabic due to the challenges and complexity of the Arabic language. We propose in this paper a comparison between traditional neural network-based models and pre-trained transformers. The experimental results show that transformers are a promising approach for the task of Arabic sarcasm detection.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"55 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121460065","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993489
Rahima Bentrcia, Meriem Tallai, Asma Mekdour
There is an immense need for recognition systems that rely on Arabic characters to provide a reliable and fast processing of data. Since Arabic writing is widely used in various real-world applications, this motivated us to develop a recognition system which recognizes mixed fonts printed Arabic letters of different sizes besides ligatures, digits, and punctuation marks. The proposed system consists of the preprocessing phase, the feature extraction phase, and the recognition phase which exploiting two models of Convolutional Neural Networks CNNs to recognize the characters. The experimental results are very promising as the second model (CNN model 2) outperforms the first model (CNN model1) and achieves an accuracy rate of 99.86%.
{"title":"A Deep Learning Approach to Recognize Mixed Fonts Printed Arabic Characters","authors":"Rahima Bentrcia, Meriem Tallai, Asma Mekdour","doi":"10.1109/ISIA55826.2022.9993489","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993489","url":null,"abstract":"There is an immense need for recognition systems that rely on Arabic characters to provide a reliable and fast processing of data. Since Arabic writing is widely used in various real-world applications, this motivated us to develop a recognition system which recognizes mixed fonts printed Arabic letters of different sizes besides ligatures, digits, and punctuation marks. The proposed system consists of the preprocessing phase, the feature extraction phase, and the recognition phase which exploiting two models of Convolutional Neural Networks CNNs to recognize the characters. The experimental results are very promising as the second model (CNN model 2) outperforms the first model (CNN model1) and achieves an accuracy rate of 99.86%.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126581631","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993601
Abdeldjalil Chougui, Achraf Moussaoui, A. Moussaoui
Detecting plant diseases is usually difficult without an experts knowledge. In this study we want to propose a new classification model based on deep learning that will be able to classify and identify different plant-leaf diseases with high accuracy that outperforms the state of the art approaches and previous works. Using only training images, CNN can automatically extract features for classification, and achieve high classification performance. We used two datasets in this study, PlantVillage dataset containing 54,303 healthy and unhealthy leaf images divided into 38 categories by species and disease, and Tomato dataset containing 11,000 healthy and unhealthy tomato leaf images with nine diseases to train the models. We propose a deep convolutional neural network architecture, with and without attention mechanism, and we tuned 4 pretrained models that have been trained on large dataset such as MobileNet, VGG-16, VGG-19 and ResNET. We also tuned 2 ViT models, the vit b32 from keras and the base patch 16 from google. Our porposed model obtained an accuracy up to 97.74%. The pretrained models gave an accuracy up to 99.52%. And the ViT models obtained an accuracy up to 99.7%. This study may aid in detecting the plant leaf diseases and improve life conditions to plants which will improve quality of humans life.
{"title":"Plant-Leaf Diseases Classification using CNN, CBAM and Vision Transformer","authors":"Abdeldjalil Chougui, Achraf Moussaoui, A. Moussaoui","doi":"10.1109/ISIA55826.2022.9993601","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993601","url":null,"abstract":"Detecting plant diseases is usually difficult without an experts knowledge. In this study we want to propose a new classification model based on deep learning that will be able to classify and identify different plant-leaf diseases with high accuracy that outperforms the state of the art approaches and previous works. Using only training images, CNN can automatically extract features for classification, and achieve high classification performance. We used two datasets in this study, PlantVillage dataset containing 54,303 healthy and unhealthy leaf images divided into 38 categories by species and disease, and Tomato dataset containing 11,000 healthy and unhealthy tomato leaf images with nine diseases to train the models. We propose a deep convolutional neural network architecture, with and without attention mechanism, and we tuned 4 pretrained models that have been trained on large dataset such as MobileNet, VGG-16, VGG-19 and ResNET. We also tuned 2 ViT models, the vit b32 from keras and the base patch 16 from google. Our porposed model obtained an accuracy up to 97.74%. The pretrained models gave an accuracy up to 99.52%. And the ViT models obtained an accuracy up to 99.7%. This study may aid in detecting the plant leaf diseases and improve life conditions to plants which will improve quality of humans life.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132180435","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993598
Tarek Amine Haddad, D. Hedjazi, Sofiane Aouag
In this work, we focus on optimizing traffic signal control at an isolated intersection and subsequently alleviate the traffic flow. We propose a new Deep Reinforcement Learning-based approach. Thus, the traffic network controller in an isolated intersection is modelled as an intelligent agent that perceives the discrete state encoding of traffic information as the network inputs. Our contribution resides to use a Double Deep Q-Network (DDQN). This argues that the idea of having a simplified state and reward formula facilitates the training of the agent by simplifying the convergence of the latter. It dynamically select the phases improving the traffic quality. The experimental results shows that the proposed approach is competitive in terms of Average Waiting Time, Average Queue Length, Average Fuel Consumption and Average Emission CO2 at intersection when compared to some baseline methods.
{"title":"A New Deep Reinforcement Learning-Based Adaptive Traffic Light Control Approach for Isolated Intersection","authors":"Tarek Amine Haddad, D. Hedjazi, Sofiane Aouag","doi":"10.1109/ISIA55826.2022.9993598","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993598","url":null,"abstract":"In this work, we focus on optimizing traffic signal control at an isolated intersection and subsequently alleviate the traffic flow. We propose a new Deep Reinforcement Learning-based approach. Thus, the traffic network controller in an isolated intersection is modelled as an intelligent agent that perceives the discrete state encoding of traffic information as the network inputs. Our contribution resides to use a Double Deep Q-Network (DDQN). This argues that the idea of having a simplified state and reward formula facilitates the training of the agent by simplifying the convergence of the latter. It dynamically select the phases improving the traffic quality. The experimental results shows that the proposed approach is competitive in terms of Average Waiting Time, Average Queue Length, Average Fuel Consumption and Average Emission CO2 at intersection when compared to some baseline methods.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124226108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993569
Leila Abbad, Djallel Brahmia, Mohamed Nadir Cherfia
In the last several decades and as a result of various kinds of man-made activities, industrialization and human urbanization, the atmospheric environment pollution became a real threat to the human's health. The particles with a diameter of less than 2.5µm, one of the most harmful pollutants present in the air as it causes diseases in the respiratory system as well as cardiovascular ones. Consequently, it is beneficial to predict the particulate matter PM2.5 concentrations with high accuracy for the purpose of to alert people to make the right decision in order to fix the situation and improve the air quality especially in environments where it is essential. The prediction of the PM2.5 concentration have to pass throw a pre-processing stage then fed to the multiple models by passing a data chunk of twelve days to get the prediction for the next day. In this article, a comparative study between different Artificial Intelligence predictions models is presented: Bidirectional Long Short-Term Memory (Bi-LSTM), Time Distributed Convolutional Neural Network (CNN), and a hybrid model combining both CNN and Bi-LSTM. For this purpose, several architectures were used for the different models: Multi Inputs - Multi Outputs, Multi Inputs - Single Output and the univariate. The CNN extracts the internal spatial correlation between variables and the Bi-LSTM extracts the temporal patterns, the hybridization process proposed of those two models with the multiple Inputs -Single Output architecture gave us the most accurate results.
{"title":"Study and comparison of machine learning models for air PM 2.5 concentration prediction","authors":"Leila Abbad, Djallel Brahmia, Mohamed Nadir Cherfia","doi":"10.1109/ISIA55826.2022.9993569","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993569","url":null,"abstract":"In the last several decades and as a result of various kinds of man-made activities, industrialization and human urbanization, the atmospheric environment pollution became a real threat to the human's health. The particles with a diameter of less than 2.5µm, one of the most harmful pollutants present in the air as it causes diseases in the respiratory system as well as cardiovascular ones. Consequently, it is beneficial to predict the particulate matter PM2.5 concentrations with high accuracy for the purpose of to alert people to make the right decision in order to fix the situation and improve the air quality especially in environments where it is essential. The prediction of the PM2.5 concentration have to pass throw a pre-processing stage then fed to the multiple models by passing a data chunk of twelve days to get the prediction for the next day. In this article, a comparative study between different Artificial Intelligence predictions models is presented: Bidirectional Long Short-Term Memory (Bi-LSTM), Time Distributed Convolutional Neural Network (CNN), and a hybrid model combining both CNN and Bi-LSTM. For this purpose, several architectures were used for the different models: Multi Inputs - Multi Outputs, Multi Inputs - Single Output and the univariate. The CNN extracts the internal spatial correlation between variables and the Bi-LSTM extracts the temporal patterns, the hybridization process proposed of those two models with the multiple Inputs -Single Output architecture gave us the most accurate results.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131080060","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993565
A. Chouchane, Mohcene Bessaoudi, A. Ouamane, Oussama Laouadi
Kinship verification from face images is a motivating field of study in computer vision, involving many researches works because-of its importance in many potential applications, such as forensics and finding missing children. This application of automatically determining whether persons share a blood re-lationship by examining their facial characteristics, i.e., features. In this work, we develop an efficient method named Hist-Gabor based on the histogram features extracted from basic Gabor wavelet in order to represent face images with high discriminate power. Indeed, we examine the use of deep features collected from a convolutional neural network model called VGG-face and shallow features by our new Gabor wavelet invoking a powerful dimensionality reduction method named Tensor Cross-view Quadratic Analysis (TXQDA). Empirically, our experiments demonstrate that the proposed approach outperforms the pre-vious state-of-the-art in the challenging datasets Cornell and TSKinFace.
{"title":"Face Kinship Verification Based VGG16 and new Gabor Wavelet Features","authors":"A. Chouchane, Mohcene Bessaoudi, A. Ouamane, Oussama Laouadi","doi":"10.1109/ISIA55826.2022.9993565","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993565","url":null,"abstract":"Kinship verification from face images is a motivating field of study in computer vision, involving many researches works because-of its importance in many potential applications, such as forensics and finding missing children. This application of automatically determining whether persons share a blood re-lationship by examining their facial characteristics, i.e., features. In this work, we develop an efficient method named Hist-Gabor based on the histogram features extracted from basic Gabor wavelet in order to represent face images with high discriminate power. Indeed, we examine the use of deep features collected from a convolutional neural network model called VGG-face and shallow features by our new Gabor wavelet invoking a powerful dimensionality reduction method named Tensor Cross-view Quadratic Analysis (TXQDA). Empirically, our experiments demonstrate that the proposed approach outperforms the pre-vious state-of-the-art in the challenging datasets Cornell and TSKinFace.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133070538","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993540
Mahdia Azzouz, Saïda Boukhedouma, Z. Alimazighi
Customer churn is one of the most critical issues faced by companies. These turn towards prediction techniques to predict the churn of their customers, because it is more expensive to acquire a new customer inside of retaining existing one. In this paper, we propose a process-based approach to detect potential customer churn and provide early warning indicator of problems that could lead to customer's loss and open up opportunities to implement effective retention strategies. The predictive churn model is determined by applying a set of data mining and machine learning algorithms, in order to keep flexible choice of the best prediction algorithm. Once the categories of churners are determined, association rule mining algorithm is applied to analyze and detect customer churn causes. The proposed approach is based on the CRISP-DM process with flexible choice of predictive model since it implements different machine learning algorithms and allows the selection of the most appropriate one for better churn prediction (the best model). The proposed approach is illustrated on a case study and the results indicate that the system is effective in detecting customer churners and addressing appropriate retention solutions.
{"title":"An approach with flexible choice of model for customer churn prediction and retention help","authors":"Mahdia Azzouz, Saïda Boukhedouma, Z. Alimazighi","doi":"10.1109/ISIA55826.2022.9993540","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993540","url":null,"abstract":"Customer churn is one of the most critical issues faced by companies. These turn towards prediction techniques to predict the churn of their customers, because it is more expensive to acquire a new customer inside of retaining existing one. In this paper, we propose a process-based approach to detect potential customer churn and provide early warning indicator of problems that could lead to customer's loss and open up opportunities to implement effective retention strategies. The predictive churn model is determined by applying a set of data mining and machine learning algorithms, in order to keep flexible choice of the best prediction algorithm. Once the categories of churners are determined, association rule mining algorithm is applied to analyze and detect customer churn causes. The proposed approach is based on the CRISP-DM process with flexible choice of predictive model since it implements different machine learning algorithms and allows the selection of the most appropriate one for better churn prediction (the best model). The proposed approach is illustrated on a case study and the results indicate that the system is effective in detecting customer churners and addressing appropriate retention solutions.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134357365","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-29DOI: 10.1109/ISIA55826.2022.9993531
Khadidja Delloul, S. Larabi
Image captioning methods come short when being used to describe scenes for the blind and visually impaired, because not only do they focus exclusively on salient objects, eliminating background and surrounding information, but they also do not offer egocentric positional descriptions of objects regarding the users, failing by that to give them the opportunity to understand and rebuild the scenes they are in. Furthermore, the majority of solutions neglect depth information, and models are trained solely on 2D (RGB) images, leading to less accurate prepositions and words or phrases' order. In this paper, we will offer the blind and visually impaired more descriptive captions for almost every region present in the image by the use of DenseCap model. Our contribution lies within the use of depth information that will be estimated by AdaBins model in order to enrich captions with positional information regarding the users, helping them understand their surroundings.
{"title":"Egocentric Scene Description for the Blind and Visually Impaired","authors":"Khadidja Delloul, S. Larabi","doi":"10.1109/ISIA55826.2022.9993531","DOIUrl":"https://doi.org/10.1109/ISIA55826.2022.9993531","url":null,"abstract":"Image captioning methods come short when being used to describe scenes for the blind and visually impaired, because not only do they focus exclusively on salient objects, eliminating background and surrounding information, but they also do not offer egocentric positional descriptions of objects regarding the users, failing by that to give them the opportunity to understand and rebuild the scenes they are in. Furthermore, the majority of solutions neglect depth information, and models are trained solely on 2D (RGB) images, leading to less accurate prepositions and words or phrases' order. In this paper, we will offer the blind and visually impaired more descriptive captions for almost every region present in the image by the use of DenseCap model. Our contribution lies within the use of depth information that will be estimated by AdaBins model in order to enrich captions with positional information regarding the users, helping them understand their surroundings.","PeriodicalId":169898,"journal":{"name":"2022 5th International Symposium on Informatics and its Applications (ISIA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129334063","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：