Year: 2024
Type: dissertation
Abstract: Over the last few decades, much research on facial expression recognition has predominantly focused on posed, static facial images, often overlooking the importance of dynamic and spontaneous information. This dissertation addresses these gaps by exploring the roles of dynamic and spontaneous aspects in emotion recognition through comprehensive reviews and empirical studies of both humans and automated systems. In the first set of studies, various expression formats - dynamic, target, and non-target static - are analysed to determine the conditions under which dynamic information significantly enhances recognisability of expressions. Results revealed that dynamic cues play a compensatory role, particularly aiding recognition when static expressions fail to represent target emotions adequately. Subsequently, Chapter 2 reviews the existing databases of spontaneous and dynamic facial expressions, detailing their conceptual, technical, and practical features, thereby providing a comprehensive benchmark for research on encoding and decoding facial expressions. Employing automated facial expression analysis tools, Chapter 3 presents an empirical cross-corpus evaluation of the databases reviewed in Chapter 2. Findings showed that, although recognition rates for spontaneous databases generally remain low, they vary significantly across databases, highlighting the inherent difficulty and variability in recognising spontaneous expressions. Furthermore, this work elucidates the critical roles of featural parameters – prototypicality, ambiguity, and complexity – in accurate emotion recognition. In sum, the findings demonstrate that dynamic properties and spontaneous aspects convey important information that significantly influences the human and machine recognition of facial expressions.
Author: Kim, Hyunwoo
Institution: University College London
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6991627048
