International Journal of Intelligent Engineering Informatics (4 papers in press)

Regular Issues

• A Review of Advancements in Deep Learning-Based Shadow Detection and Removal in Image and Video Analysis  
  by Hui Liu, Kin Sam Yen 
  Abstract: Shadow detection and removal are vital in image processing and computer vision applications, and have impacted diverse industries such as automated driving, medicine, and agriculture. Shadows in images and videos can significantly impede algorithm performance. Numerous models and techniques have been proposed to address this concern. This study presents a comprehensive review and analysis of research focusing on shadow detection and removal, including image and video-based algorithms related to deep learning (DL) approaches since 2017. The aim is to explore the latest advancements and developments in integrating deep learning approaches in this field. The architecture of the DL models and their performance are analysed. A noticeable trend is observed as shadow detection and removal algorithms transition from conventional image processing and analysis methods to DL approaches. The challenges, such as data scarcity, training paradigms, incorporation of temporal information, and utilisation of advanced models, remain open research areas.
  Keywords: shadow detection; shadow removal; deep learning; shadow image analysis; video analysis.
  DOI: 10.1504/IJIEI.2024.10062950
   
  
• SwinRelTR: An Efficient Single-Stage Scene Graph Generation Model for Low-Resolution Images  
  by Mohammad Essam, Howida A. Shedeed, Mohamed F. Tolba, Dina Khattab 
  Abstract: Targeting low-resolution imagery is crucial in democratising computer vision technologies, facilitating applicability in resource-limited environments where high-resolution data is often unreachable. Scene graphs have proven to be a powerful representation for capturing the hierarchical relationships between objects in an input image, providing a structured visual scene understanding. Nevertheless, all scene graph generation models focus on high-resolution images, neglecting the challenges posed by low-resolution images. This paper presents a novel approach called SwinRelTR for generating scene graphs designed specifically for low-resolution images. The proposed model addresses the limitations associated with low-resolution images by utilising the Swin transformer as a backbone instead of the convolution neural network in the original RelTR model. The Visual Genome dataset is utilised to compare the SwinRelTR results with the state-of-the-art approaches. It has been proven that this approach outperforms several state-of-the-art approaches as well as the original RelTR model on low-resolution images.
  Keywords: low-resolution; scene graph; scene graph generation; SGG; visual scene understanding; visual relationship detection.
  DOI: 10.1504/IJIEI.2024.10063131
   
  
• Exploring CNN-based Transfer Learning Approaches for Arabic Alphabets Sign Language Recognition using the ArSL2018 Datase  
  by Houssem Lahiani, Mondher Frikha 
  Abstract: Arabic alphabets sign language (ArASL) recognition is an important topic that has gotten insufficient attention Regardless of its significance in the Arab world. This research compares CNN-based transfer learning models for Arabic alphabets sign language (ArASL) recognition using the ArSL2018 dataset, which comprises 54,049 pictures representing 32 sign and letter classes. Three pre-trained models are examined (InceptionV3, VGG16, and MobileNetV2) and compared using a training and evaluation dataset split. We use transfer learning to fine-tune these models on the ArSL2018 dataset and compare their performance. Our experimental findings indicate that the MobileNetV2 model exceeds the other models in terms of accuracy, achieving an overall accuracy of 96%, which exceeds the state-of-the-art results, reported in previous works. Our study demonstrates that transfer learning is an effective approach for recognising Arabic alphabets sign language using CNN-based models and provides insights into the suitability of different pre-trained models for this task.
  Keywords: convolutional neural network; CNN; HMI; transfer learning; Arabic alphabets sign language; ArASL.
  DOI: 10.1504/IJIEI.2024.10063260
   
  
• Android Malware Detection for Timely Detection Using Multi-Class Deep Learning Methods  
  by Anusha Muthukrishnan, Karthika M 
  Abstract: Android malware has emerged as a severe danger to national security because of the widespread usage of smartphones and the inherent risk it provides to its users. Due to code obfuscation, antivirus products and other typical detection algorithms struggle to catch Android malware, which has increased. Deep learning-based solutions safeguard legitimate Android users against fraudulent apps, which is a must. The methods categorise Android malware using multiple feature representations. Unfortunately, as more apps use classifiers, the temptation to weaken them develops. According to current research, deep learning is being used to identify malware. A learning-based classifier processes Android application properties to test deep learning for Android virus detection safety (apps). Considering the features' importance to the classification problem and the costs of changing them, we suggest an encoder-decoder-based CNN feature selection approach to make the classifier tougher to bypass. We also provide a spider monkey optimisation-based Bi-LSTM method that combines classifiers from our feature selection strategy to improve system security without compromising detection accuracy. Testing on CICInv and Mal2021/CICInv sample sets proved the suggested strategys efficacy against malicious Android malware attacks. In addition, any malware detection setup can employ our secure-learning paradigm.
  Keywords: Android malware; encoder-decoder based CNN; spider monkey optimisation; Bi-LSTM; Machine Learning; Deep Learning; Detection Accuracy; Encoder-Decoder-Based.
  DOI: 10.1504/IJIEI.2024.10063350