Phishing attacks are a growing threat that has evolved along with technological advancements. Existing detection methods struggle with constantly evolving tactics and "zero-day" attacks that exploit unknown vulnerabilities. This paper proposes a novel approach for identifying phishing web pages built on phishing kits and, consequently, detecting such attack attempts. Our method, Phish Fighter, analyzes the HTML code structure, focusing on recurring blocks across different phishing pages derived from the same source. These features are then fed to clustering and classification components to detect common structural patterns without relying on textual or visual content. This approach overcomes the limitations of existing solutions, and is robust against attacks that target various brands. Furthermore, we implemented a dedicated module for continuous data updates for the Phish Fighter. This module effectively recognizes even "zero-day" phishing attempts by analyzing only three pages associated with a new phishing kit. In addition, we successfully identified the phishing pages created by cloning the original source code of legitimate entities without requiring prior knowledge to distinguish such clones. The results support the efficiency and accuracy of this approach: weighted precision, recall and F1 score are all greater than 90%, and the respective micro-averaged metrics are above 95%.

Disinformation in the form of news articles, also called fake news, is used by multiple actors for nefarious purposes, such as gaining political advantages. A key component for fake news detection is the ability to find similar articles in a large documents corpus, for tracking narrative changes and identifying the root source (patient zero) of a particular piece of information. This paper presents new techniques based on textual and semantic similarity that were adapted for achieving this goal on large datasets of news articles. The aim is to determine which of the implemented text similarity techniques is more suitable for this task. For text similarity, a Locality-Sensitive Hashing is applied on n-grams extracted from text to produce representations that are further indexed to facilitate the quick discovery of similar articles. The semantic textual similarity technique is based on sentence embeddings from pre-trained language models, such as BERT, and Named Entity Recognition. The proposed techniques are evaluated on a collection of Romanian articles to determine their performance in terms of quality of results and scalability. The presented techniques produce competitive results. The experimental results show that the proposed semantic textual similarity technique is better at identifying similar text documents, while the Locality-Sensitive Hashing text similarity technique outperforms it in terms of execution time and scalability. Even if they were evaluated only on Romanian texts and some of them are based on pre-trained models for the Romanian language, the methods that are the basis of these techniques allow their extension to other languages, with few to no changes, provided that there are pre-trained models for other languages as well. As for a cross-lingual setup, more changes are needed along with tests to demonstrate this capability. Based on the obtained results, one may conclude that the presented techniques are suitable to be integrated into a decentralized anti-disinformation platform for fact-checking and trust assessment.

Assistive devices developed for visually impaired users help them understand the environment so they can navigate with minimum additional help. Such devices greatly rely on image processing techniques to distinguish essential features from the input images. The ground surface represents one key element that needs to be detected. The user can move freely on the ground surface but also obstacles are usually located on this surface. As assistive devices need to provide information in real time, any designed algorithm must achieve real time results. In this paper, we present a novel algorithm to detect the ground plane in video sequences. The sequences are acquired by a stereo camera worn by a visually impaired user. Our algorithm uses a V-disparity map approach and it is tailored for the challenges that arise from the fact that the stereo camera has six degrees-of-freedom. We evaluate the accuracy of our algorithm using multiple real life scenarios. An evaluation of the performances of the algorithm was also performed to prove its accuracy and speed. We also discuss a possible speed up of the proposed algorithm but at a minor cost of accuracy.