Title: Detecting violence in images within the context of protests using deep learning
Type of Project: Master thesis
With the continuous increase of computing processing, deep learning is gaining popularity in a variety of complex tasks such as image and natural language processing previously dominated by rule-based algorithms.
Deep learning is a class of algorithms inspired by on the information processing patterns found in biological neuron systems such as the human brain. It has proved successful at learning underlying patterns in a wide variety of areas where large amount of data is available.
In terms computer vision, deep learning has become the new state-of-the-art in terms of labeling images into discrete categories. However, these categories tend to be relatively simple, usually referring to objects. Thus, there is room to explore the power of these models when classifying more abstract and complex concepts such as violence.
Based on that, this project will explore the capabilities of using deep learning to classify images as violent or non-violent in the context of protests. Focusing on protests provides a more meaningful scope for identifying the applicability of classifying violence, while offering to be more approachable than considering all environments that such behaviour may occur. Moreover, this project is tied to a broader research effort that investigates the influence of violence on how stories propagates on social media that could benefit from having an automated labeling tool. Specifically we will consider the following question:
How well can deep learning perform as to differentiate between violent and nonviolent images within the context of protests
Jesper Henrichsen (email@example.com)
Lucas Klafke Beck (firstname.lastname@example.org)
Luca Rossi (email@example.com) – Supervisor
Zeljko Agic (firstname.lastname@example.org)- Supervisor