Image Intelligence (Sweden)

The incel subculture has gained increasing attention due to its toxic nature and its association with real-world violence. This paper investigates the prevalence and characteristics of violent threatful communication within incel forums, focusing on a platform known as Blackpill. We have trained a machine learning model to detect violent threatful language and analyzed the posts. The analysis concentrated on three key aspects: the identity of perpetrators (categorized into first-person, third-person, or generalized), the targets (individuals, groups, or general targets), and the types of violence described (general violence, sexual violence, self-harm, and military violence). The analysis showed that the most common type violent threatful communication involved generalized perpetrators targeting groups. Additionally, 13.5% of the violent threatful communication contained coded language, including references to video games to obscure violent intentions. A smaller proportion of the posts (4.1%) glorified past mass shooters and violent criminals.This research highlights the complexities of identifying violent rhetoric in online forums and the use of coded language to evade detection, emphasizing the need for refined models in threat detection.

One of the most challenging threats to the security of society is attacks from violent lone offenders. Identifying potential offenders is difficult since they act alone and do not necessarily communicate with others. However, several targeted violent attacks have been preceded by communication published on social media and the internet. Such communication is a valuable component when conducting risk and threat assessments.In this paper, we introduce a diagnostic measure of the risk of violent behavior based on text analysis. Using automated text analysis, we extract psychological variables and warning indicators from a given text and summarize these in an index that we denote as the general risk index. When developing the general risk index, we analyzed data (text) from 208 288 users on 32 online environments with diverse ideologies/orientations, including 76 previous violent lone offenders. A receiver operating characteristics (ROC) analysis showed that, when using the general risk index, it was possible to correctly classify between 90% and 96% of the cases depending on the comparison sample. These results support the predictive validity of the general risk index, suggesting that the risk index can be used to identify individuals with an increased risk of committing violent attacks that need further investigation.

Sagittal synostosis is the most common type of premature suture closure, and many surgical techniques are used to correct scaphocephalic skull shape. Given the rarity of direct comparisons of different surgical techniques for correcting craniosynostosis, this study compared outcomes of craniotomy combined with springs and H-craniectomy for nonsyndromic sagittal synostosis. Comparisons were performed using available preoperative and postoperative imaging and follow-up data from the 2 craniofacial national referral centers in Sweden, which perform 2 different surgical techniques: craniotomy combined with springs and H-craniectomy (the Renier technique). The study included 23 pairs of patients matched for sex, preoperative cephalic index, and age. Cephalic index, total intracranial volume (ICV), and partial ICV were measured before surgery and at 3 years of age, with volume measurements compared against those of preoperative and postoperative controls. Perioperative data included operation time, blood loss, volume of transfused blood, and length of hospital stay. Craniotomy combined with springs resulted in less bleeding and lower transfusion rates than H-craniectomy. Although the spring technique requires 2 operations, the mean total operation time was similar for the methods. Of the 3 complications that occurred in the group treated with springs, 2 were spring-related. The compiled analysis of changes in cephalic index and partial volume distribution revealed that craniotomy combined with springs resulted in superior morphologic correction. The findings showed that craniotomy combined with springs normalized cranial morphology to a greater extent than H-craniectomy based on changes in cephalic index and total and partial ICVs over time. Therapeutic, III.

We propose a method for multimodal deformable image registration which combines a powerful deep learning approach to generate CoMIRs, dense image-like representations of multimodal image pairs, with INSPIRE, a robust framework for monomodal deformable image registration. We introduce new equivariance constraints to improve the consistency of CoMIRs under deformation. We evaluate the method on three publicly available multimodal datasets: remote sensing, histological, and cytological. The proposed method demonstrates general applicability and consistently outperforms reference registration tools elastix and VoxelMorph. We share source code of the proposed method and complete experimental setup as open-source at: https://github.com/MIDA-group/CoMIR_INSPIRE.

Many problems in applied computer science can be expressed in a graph setting and solved by finding an appropriate vertex labeling of the associated graph. It is also common to identify the term “appropriate labeling” with a labeling that optimizes some application-motivated objective function. The goal of this work is to present two algorithms that, for the objective functions in a general format motivated by image processing tasks, find such optimal labelings. Specifically, we consider a problem of finding an optimal binary labeling for the objective function defined as the max-norm over a set of local costs of a form that naturally appears in image processing. It is well known that for a limited subclass of such problems, globally optimal solutions can be found via watershed cuts, that is, by the cuts associated with the optimal spanning forests of a graph. Here, we propose two new algorithms for optimizing a broader class of such problems. The first algorithm, that works for all considered objective functions, returns a globally optimal labeling in quadratic time with respect to the size of the graph (i.e., the number of its vertices and edges) or, for an image associated graph, the size of the image. The second algorithm is more efficient, with quasi-linear time complexity, and returns a globally optimal labeling provided that the objective function satisfies certain given conditions. These conditions are analogous to the submodularity conditions encountered in max-flow/min-cut optimization, where the objective function is defined as sum of all local costs. We will also consider a refinement of the max-norm measure, defined in terms of the lexicographical order, and examine the algorithms that could find minimal labelings with respect to this refined measure.

[This corrects the article DOI: 10.3389/fphys.2016.00330.].

Capturing and characterizing collective human activities in a geographic space have become much easier than ever before in the big era. In the past few decades it has been difficult to acquire the spatiotemporal information of human beings. Thanks to the boom in the use of mobile devices integrated with positioning systems and location-based social media data, we can easily acquire the spatial and temporal information of social media users. Previous studies have successfully used street nodes and geo-tagged social media such as Twitter to predict users’ activities. However, whether human activities can be well represented by social media data remains uncertain. On the other hand, buildings or architectures are permanent and reliable representations of human activities collectively through historical footprints. This study aims to use the big data of US building footprints to investigate the reliability of social media users for human activity prediction. We created spatial clusters from 125 million buildings and 1.48 million Twitter points in the US. We further examined and compared the spatial and statistical distribution of clusters at both country and city levels. The result of this study shows that both building and Twitter data spatial clusters show the scaling pattern measured by the scale of spatial clusters, respectively, characterized by the number points inside clusters and the area of clusters. More specifically, at the country level, the statistical distribution of the building spatial clusters fits power law distribution. Inside the four largest cities, the hotspots are power-law-distributed with the power law exponent around 2.0, meaning that they also follow the Zipf’s law. The correlations between the number of buildings and the number of tweets are very plausible, with the r square ranging from 0.53 to 0.74. The high correlation and the similarity of two datasets in terms of spatial and statistical distribution suggest that, although social media users are only a proportion of the entire population, the spatial clusters from geographical big data is a good and accurate representation of overall human activities. This study also indicates that using an improved method for spatial clustering is more suitable for big data analysis than the conventional clustering methods based on Euclidean geometry.

Abstract. Calving is an important process in glacier systems terminating in the ocean and more observations are needed to improve our understanding of the undergoing processes and be able to parameterise calving in larger scale models. Time-lapse cameras are good tools for monitoring calving fronts of glaciers and they have been used widely where conditions are favourable. However, automatic image analysis to detect and calculate the size of calving events has not been developed so far. Here, we present a method that fills this gap using image analysis tools. First, the calving front is segmented. Second, changes between two images are detected and a mask is produced to delimit the calving event. Third, we calculate the area given the front and camera positions as well as camera characteristics. To illustrate our method, we analyse two image time series from two cameras placed at different locations in 2014 and 2015 and compare the automatic detection results to a manual detection. We find a good match when the weather is favorable but the method fails with dense fog or high illumination conditions. Furthermore, results show that calving events are more likely to occur (i) close to where subglacial melt water plumes have been observed to rise at the front and (ii) close to one another.