Computer
Raja’a M. Mohammed; Suhad M. Kadhem
Abstract
Sign language (SL) is Non-verbal communication and a way for thedeaf and mute to communicate without words. A deaf and mute person's hands,face, and body shows what they want to say. Since the number of deaf and dumbpeople is increasing, there must be other ways to learn sign language orcommunicate with ...
Read More ...
Sign language (SL) is Non-verbal communication and a way for thedeaf and mute to communicate without words. A deaf and mute person's hands,face, and body shows what they want to say. Since the number of deaf and dumbpeople is increasing, there must be other ways to learn sign language orcommunicate with deaf and dumb people. One of these ways is using advancedtechnology to produce systems that help the deaf/dumb, such as creatingrecognition and sign language translators. This paper presents an applicationthat works on the computer for machine translation of Iraqi sign language intwo directions from sign language to Arabic language (text/speech) and fromArabic language(text) to Iraqi sign language. The proposed system uses aConvolution Neural Network (CNN) to classify sign language based on itsfeatures to predicate the sign meaning. The sign language to Arabiclanguage(text/speech) part of the proposed system has an accuracy of 99.3% forletters.
Computer
Asmaa Hasan Alrubaie; Maisa'a Abid Ali Khodher; Ahmed Talib Abdulameer
Abstract
Target detection, one of the key functions of computer vision, has grown in importance as a study area over the past two decades and is currently often employed. In a certain video, it seeks to rapidly and precisely detect and locate a huge amount of the objects according to redetermined categories. ...
Read More ...
Target detection, one of the key functions of computer vision, has grown in importance as a study area over the past two decades and is currently often employed. In a certain video, it seeks to rapidly and precisely detect and locate a huge amount of the objects according to redetermined categories. The two forms of deep learning (DL) algorithms that are used in the model training algorithm are single-stage and 2-stage algorithms of detection. The representative algorithms for every level have been thoroughly discussed in this work. The analysis and comparison of numerous representative algorithms in this subject is after that explained. Last but not least, potential obstacles to target detection are anticipated.
Computer
Suha Mohammed Saleh; Abdulamir A. Karim
Abstract
From big data analytics to computer vision and human-level control, deep learning has been effectively applied to a wide range of complicated challenges. However, these same deep learning advancements have also been used to develop malicious software that threatens individuals' personal data, democratic ...
Read More ...
From big data analytics to computer vision and human-level control, deep learning has been effectively applied to a wide range of complicated challenges. However, these same deep learning advancements have also been used to develop malicious software that threatens individuals' personal data, democratic processes, and even national security. Apps backed by deep learning have lately appeared, with deepfake being one of the most notable. Deepfake algorithms can create fake images and videos that humans cannot distinguish them from authentic ones. One of the fields that deep learning accomplished major success is face synthesis and animation generation. On the other hand, it can create unethical software called deepfake that presents a severe privacy threat or even a huge security risk that can affect innocent people. This work introduces the most recent algorithms and methods used in deepfake. In addition, it provides a brief explanation of the principles that underpin these technologies and facilitates the development of this field by identifying the challenges and scopes that require further investigation in the future.
Computer
Sabah Abdulazeez Jebur; Khalid A. Hussein; Haider Kadhim Hoomod
Abstract
The use of video surveillance systems has increased due to security concerns and their relatively low cost. Researchers are working to create intelligent Closed Circuit Television (CCTV) cameras that can automatically analyze behavior in real-time to detect anomalous behaviors and prevent dangerous accidents. ...
Read More ...
The use of video surveillance systems has increased due to security concerns and their relatively low cost. Researchers are working to create intelligent Closed Circuit Television (CCTV) cameras that can automatically analyze behavior in real-time to detect anomalous behaviors and prevent dangerous accidents. Deep Learning (DL) approaches, particularly Convolutional Neural Networks (CNNs), have shown outstanding results in video analysis and anomaly detection. This research paper focused on using Inception-v3 transfer learning approaches to improve the accuracy and efficiency of abnormal behavior detection in video surveillance. The Inceptionv3 network is used to classify keyframes of a video as normal or abnormal behaviors by utilizing both pre-training and fine-tuning transfer learning approaches to extract features from the input data and develop a new classifier. The UCF-Crime dataset is used to train and evaluate the proposed models. The performance of both models was evaluated using accuracy, recall, precision, and F1 score. The fine-tuned model achieved 88.0%, 89.24%, 85.83%, and 87.50% for these measures, respectively. In contrast, the pre-trained model obtained 86.2%, 86.43%, 84.62%, and 85.52%, respectively. These results demonstrate that transfer learning using Inception-v3 architecture can effectively classify normal and abnormal behaviors in videos, and fine-tuning the weights of the layers can further improve the model's performance.
Computer
Afrah Salman Dawood; Zena Mohammed Faris
Abstract
Recently, Deep Learning (DL) has accomplished enormous prosperity in various areas, like natural language processing (NLP), image processing, different medical issues and computer vision. Both Machine Learning (ML) and DL as compared to traditional methods, can learn and make better and enhanced use ...
Read More ...
Recently, Deep Learning (DL) has accomplished enormous prosperity in various areas, like natural language processing (NLP), image processing, different medical issues and computer vision. Both Machine Learning (ML) and DL as compared to traditional methods, can learn and make better and enhanced use of datasets for feature extraction. This paper is divided into three parts. The first part introduces a detailed information about different characteristics and learning types in terms of learning problems, hybrid learning problems, statistical inference and learning techniques; besides to an exhausted historical background about feature learning and DL. The second part is about the major architectures of DL with mathematical equations and clarified examples. These architectures include Autoencoders (AEs), Generative Adversarial Networks (GANs), Deep Belief Networks (DBNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and Recursive Neural Networks. The third part of this work represents an overview with detailed explanation about different applications and use-cases. Finally, the fourth part is about hardware/ software tools used with DL.
Computer
Muna Khalaf; Ban N. Dhannoon
Abstract
Semantic segmentation refers to labeling each pixel in the scene to its belonging object. It is a critical task for many computer vision applications that requires scene understanding because It attempts to mimic human perceptual grouping. Despite the unremitting efforts in this field, it is still a ...
Read More ...
Semantic segmentation refers to labeling each pixel in the scene to its belonging object. It is a critical task for many computer vision applications that requires scene understanding because It attempts to mimic human perceptual grouping. Despite the unremitting efforts in this field, it is still a challenge and preoccupies of researchers. Semantic segmentation performance improved using deep learning rather than traditional methods. Semantic segmentation based on deep learning models requires capturing local and global context information, where deep learning models usually can extract one of them but is challenging to integrate between them. Deep learning based on attention mechanisms can gather between the capturing of local and glopal information, so it is increasingly employed in semantic segmentation. This paper gives an introductory survey of the rising topic attention mechanisms in semantic segmentation. At first, it will discuss the concept of attention and its integration with semantic segmentation requirements. Then, it will review deep learning based on attention mechanisms in semantic segmentation.
Computer
Wildan J. Jameel; Suhad M. Kadhem; Ayad R. Abbas
Abstract
The main reason for the emergence of a deepfake (deep learning and fake) term is the evolution in artificial intelligence techniques, especially deep learning. Deep learning algorithms, which auto-solve problems when giving large sets of data, are used to swap faces in digital media to create fake media ...
Read More ...
The main reason for the emergence of a deepfake (deep learning and fake) term is the evolution in artificial intelligence techniques, especially deep learning. Deep learning algorithms, which auto-solve problems when giving large sets of data, are used to swap faces in digital media to create fake media with a realistic appearance. To increase the accuracy of distinguishing a real video from fake one, a new model has been developed based on deep learning and noise residuals. By using Steganalysis Rich Model (SRM) filters, we can gather a low-level noise map that is used as input to a light Convolution neural network (CNN) to classify a real face from fake one. The results of our work show that the training accuracy of the CNN model can be significantly enhanced by using noise residuals instead of RGB pixels. Compared to alternative methods, the advantages of our method include higher detection accuracy, lowest training time, a fewer number of layers and parameters.
Computer
Muna Ghazi; Matheel Abdulmunim
Abstract
Text summarization can be utilized for variety type of purposes; one of them for summary lecture file. A long document expended long time and large capacity. Since it may contain duplicated information, more over, irrelevant details that take long period to access relevant information. Summarization ...
Read More ...
Text summarization can be utilized for variety type of purposes; one of them for summary lecture file. A long document expended long time and large capacity. Since it may contain duplicated information, more over, irrelevant details that take long period to access relevant information. Summarization is a technique which provides the primary points of the whole document, and in the same time it will indicates the majority of the information in a small amount of time. For this reason it can save user time, decrease storage, and increase transfer speed to transmit through the internet. The summarization process will eliminate duplicated data, unimportant information, and also replace complex expression with simpler expression. The proposed method is using convolutional recurrent neural network deep model as a method for abstractive text summarization of lecture file that will be great helpful to students to address lecture notes. This method proposes a novel encoder-decoder deep model including two deep model networks which are convolutional and recurrent. The encoder part which consists of two convolutional layers followed by three recurrent layers of type bidirectional long short term memory. The decoder part which consists of one recurrent layer of type long short term memory. And also using attention mechanism layer. The proposed method training using standard CNN/Daily Mail dataset that achieved 92.90% accuracy.
Computer
Mohammed Abduljabbar Ali; Abir Jaafar Hussain; Ahmed T. Sadiq
Abstract
Crowd detection has various applications nowadays. However, detecting humans in crowded circumstances is difficult because the features of different objects conflict, making cross-state detection impossible. Detectors in the overlapping zone may therefore overreact. The proposal uses the YOLO v5 (You ...
Read More ...
Crowd detection has various applications nowadays. However, detecting humans in crowded circumstances is difficult because the features of different objects conflict, making cross-state detection impossible. Detectors in the overlapping zone may therefore overreact. The proposal uses the YOLO v5 (You Only Look Once) method to improve crowd recognition and counting. This algorithm is entirely accurate and detects things in real-time. The idea relies on edge enhancement and pre-processing to solve overlapping feature regions in the image and improve performance. The CrowdHuman data set is used to train YOLO v5. The system counts the number of humans in the image to detect a crowd. Before training, this model enhanced the image with several filters. The YOLO v5 algorithm distinguishes a person inside a crowd by utilizing the surrounding box on the head and overall body. Therefore, the number of head detection is x- coordinated compared to the body. Assume the detected heads outnumber the bodies. A square of the head will be extracted, but not a body square. Also, cropping the image reduces interference between human beings and enhances the edge features. Thus, YOLOv5 can detect it. The idea improves head and body detection by 2.17 and 4.1 percent, respectively.
sufyan zaben; Akbas Ezaldeen Ali
Abstract
Covid-19 is a deadly virus that has spread worldwide, causing millions of deaths. Chest X-ray is one of the most common methods of diagnosing the infection of Covid - 19. Therefore, this paper has presented an efficient method to detect Covid-19 through X-rays of the chest area through a Neural convolution ...
Read More ...
Covid-19 is a deadly virus that has spread worldwide, causing millions of deaths. Chest X-ray is one of the most common methods of diagnosing the infection of Covid - 19. Therefore, this paper has presented an efficient method to detect Covid-19 through X-rays of the chest area through a Neural convolution network (CNN). the proposed system has used a convolution neural network to classify the extracted features. Since CNN needs a set of data defined for training and testing, the proposed method used a public dataset of 350 pneumonia x-ray images, 300 viral images, and 350 normal images for evaluation. Besides, the proposed work achieved a satisfactory accuracy of 95% based on the X-ray image.
Zainab Mohammed Resan; Muayad Sadik Croock
Abstract
Robust and accurate indoor localization has been the goal of several researchefforts over the past decade. In the building where the GPS is not available, this projectcan be utilized. Indoor localization based on image matching techniques related to deeplearning was achieved in a hard environment. So, ...
Read More ...
Robust and accurate indoor localization has been the goal of several researchefforts over the past decade. In the building where the GPS is not available, this projectcan be utilized. Indoor localization based on image matching techniques related to deeplearning was achieved in a hard environment. So, if it wanted to raise the precision ofindoor classification, the number of image dataset of the indoor environment should be aslarge as possible to satisfy and cover the underlying area. In this work, a smartphonecamera is used to build the image-based dataset of the investigated building. In addition,captured images in real time are taken to be processed with the proposed model as a testset. The proposed indoor localization includes two phases the first one is the offlinelearning phase and the second phase is the online processing phase. In the offline learningphase, here we propose a convolutional neural network (CNN) model that sequences thefeatures of image data from some classis's dataset composed with a smartphone camera.In the online processing phase, an image is taken by the camera of a smartphone in real–time to be tested by the proposed model. The obtained results of the prediction can appointthe expected indoor location. The proposed system has been tested over variousexperiments and the obtained experimental results show that the accuracy of the predictionis almost 98.0%.
