Deep neural networks are increasingly employed to create adversarial face images, aiming to deceive face recognition systems. While the majority of studies concentrate on digital attacks, their relevance extends to face spoofing. Notably, they have the capability to generate potential face images of victims when attackers lack knowledge about individuals registered in face recognition systems. Regrettably, recent advances in attacking face recognition systems using deep neural networks, their performance, and their transferability to physical attacks (deep face spoofing) lack systematic exploration. This paper addresses this gap by presenting the first comprehensive survey of current research in this domain. The review initiates with the definition of the deep face spoofing concept and introduces a pioneering taxonomy to systematically consolidate recent advances towards deep face spoofing. The main section of the paper provides in-depth evaluations of the mechanism, performance, and applicability of diverse deep neural network-based attacking algorithms against face recognition systems. Subsequently, the paper outlines current challenges in deep face spoofing, including the absence of evaluations of recent attacks against state-of-the-art face anti-spoofing algorithms and the limited transferability of recent digital attacks to physical attacks. This part also covers open challenges in deep face spoofing detection since it is crucial to note that studying various deep face spoofing algorithms should always be seen as an effort to investigate the vulnerability of face recognition systems against such evolved attacks, and not as an endeavor to gain access for illegal purposes. To enhance accessibility to a broad range of research papers in this area, an accompanying web page ( https://github.com/dhimasarief/DFS_DFAS ) has been established. This serves as a dynamic repository of studies focusing on deep face spoofing, continuously curated with new findings and contributions.
1 June 2024 in IAES International Journal of Artificial Intelligence (IJ-AI)
To generate a machine learning (ML) and deep learning (DL) architecture with good performance, we need a decent dataset for the training and testing phases of the development process. Starting with the knowledge discovery and data mining (KDD) Cup 99 dataset, numerous datasets have been produced since 1998 to be utilized in the ML and DL-based intrusion detection systems (IDS) training and testing process. Because there are so many datasets accessible, it might be challenging for researchers to choose which dataset to employ. Therefore, a framework for evaluating dataset appropriateness with the research to be conducted is becoming increasingly crucial as new datasets are regularly created. Additionally, given the growing popularity of internet of things (IoT) devices and an increasing number of specific datasets for IoT in recent years, it is essential to have a specific framework for IoT datasets. Therefore, this research aims to develop a new framework for evaluating IoT datasets for ML and DL-based IDS. The study's findings include, first, a novel framework for assessing IoT datasets, second, a comparison of this novel framework to other existing frameworks, and third, an analysis of five IoT datasets by using the new framework.
Malaria is infectious tropical disease found in tropical countries, caused by unicellular protozoan parasite. Microscopic based diagnosis of malaria is conducted by manually examining a thin blood smear that is acquired from the infected patients. This method requires a trained human interaction and therefore it is time consuming and prone to errors. Computer Aided Diagnostics (CAD) for Malaria was developed to speed up the diagnosis and maintaining the accuracy. The experimental results showed that the system is able to recognize the infected red blood cells, the species and the life phase of the infecting Plasmodium. However, the average runtime of the detection using the original version is 41.45 seconds per image, which is too long if it will be used in the field. By measuring the runtime of each process in the program, optimization can be done by re-writing or substituting the algorithm that causes the longest runtime. Four modifications are proposed: Connected Component Labelling with Bounding Box, Contour Tracing Labelling without Inner Contour Extraction, Downscaled Clump Splitting, and Concave Point Based Clump Splitting. The experimental results showed that the system has an average runtime of 1.73 seconds while maintaining the same level of accuracy compared to the original one.
In order to support research in drug development, there is a need to classify fungi based on their genus. The first step in this research is to perform image segmentation on various fungi images. This study compares three segmentation methods with image datasets of microorganisms of fungi species from three genera: Aspergillus, Trichoderma, and Cladosporium. Otsu thresholding, adaptive thresholding, and k-means clustering are the three segmentation methods used. The comparison is evaluated using Dice and Jaccard similarity. The evaluation result shows that the adaptive thresholding method obtained the highest value with an average Jaccard score of 0.6102 and a Dice score of 0.7321. The Otsu thresholding method obtained an average Jaccard and Dice score of 0.3738 and 0.4625. Meanwhile, the k-means clustering method got an average Jaccard and Dice score of 0.2524 and 0.3272.
Intrusion Detection Systems (IDSs) remain essential for securing modern network infrastructures, where traffic data are often high-dimensional and contain redundant or weakly informative attributes. This study proposes a hybrid feature selection approach that combines Information Gain with L1-regularized selection to construct a compact and informative representation of the UNSW-NB15 dataset. The method applies relevance-based filtering followed by sparsity-driven refinement within a leakage-aware pipeline, in which preprocessing and feature selection are derived exclusively from the training data. Under a reduced six-class configuration, the proposed approach reduces 42 candidate predictors to 21 traffic-related features. Across multiple classifiers, Random Forest + IGL1 achieved the best performance, with an accuracy of 0.8432 and an F1-score of 0.8376, while MLP and Gradient Boosting also remained competitive. These findings indicate that the selected features preserve consistent discriminative patterns rather than favoring a single classifier. Overall, the study highlights the importance of leakage-aware evaluation for producing reliable, reproducible intrusion detection results. Future work will extend the analysis to the full multi-class setting and examine applicability in real-time or streaming environments.
The Face Recognition system has a challenge when the conditions of the input images have differences in quality from the images that have been enrolled in the database. One of the causes is the variation in lighting that causes illumination in image. We used images with normal lighting, as well as four images that have variations in the lighting/illumination directions. Face Image Quality Assessment (FIQA) helps the face recognition system to ensure the optimum captured image quality for enrollment and verification process. We use both supervised (FaceQnet) and unsupervised (SDD-FIQA, SER-FIQ) FIQA method against the Asian Face Image dataset. The result shows that filtering images using FIQA method can reduce FNMR by 58.89% in matching images whose light direction is from below. Images with type 2 illumination, where an image whose light comes from below matched with normal image, gave the lowest result in FRR compared to other types of illumination when tested with 3 FIQA methods.