Thesis abstract:

«p»With the rapid advancement of deep learning, computer vision has significantly evolved, expanding its applications across various domains. However, traditional frame-based cameras face limitations in dynamic scenes, such as motion blur and latency, which event cameras can potentially overcome due to their high temporal resolution and low power consumption. This research explores two key areas of computer vision: computational aesthetics in 2D vision and event-based 3D reconstruction in 3D vision. In 2D vision, we aim to develop deep learning-based methods for image aesthetic assessment and enhancement, addressing the challenges of personalised aesthetic evaluation and real-time enhancement techniques. In 3D vision, we focus on event-based 3D reconstruction, improving the accuracy and efficiency of event-driven 3D scene reconstruction through innovative event representation techniques and deep learning models. Our research seeks to advance these fields by integrating state-of-the-art methodologies, bridging the gap between traditional and event-driven vision systems, and proposing novel solutions that enhance both perceptual quality and computational efficiency.«/p»