Use this URL to cite or link to this record in EThOS:
Title: Soft shadow removal and image evaluation methods
Author: Gryka, M.
ISNI:       0000 0004 8503 214X
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
High-level image manipulation techniques are in increasing demand as they allow users to intuitively edit photographs to achieve desired effects quickly. As opposed to low-level manipulations, which provide complete freedom, but also require specialized skills and significant effort, high-level editing operations, such as removing objects (inpainting), relighting and material editing, need to respect semantic constraints. As such they shift the burden from the user to the algorithm to only allow a subset of modifications that make sense in a given scenario. Shadow removal is one such high-level objective: it is easy for users to understand and specify, but difficult to accomplish realistically due to the complexity of effects that contribute to the final image. Further, shadows are critical to scene understanding and play a crucial role in making images look realistic. We propose a machine learning-based algorithm that works well with soft shadows, that is shadows with wide penumbrae, outperforming previous techniques both in performance and ease of use. We observe that evaluation of such a technique is a difficult problem in itself and one that is often not considered throughly in the computer graphics and vision communities, even when perceptual validity is the goal. To tackle this, we propose a set of standardized procedures for image evaluation as well as an authoring system for creation of image evaluation user studies. In addition to making it possible for researchers, as well as the industry, to rigorously evaluate their image manipulation techniques on large numbers of participants, we incorporate best practices from the human-computer intraction (HCI) and psychophysics communities and provide analysis tools to explore the results in depth.
Supervisor: Brostow, G. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available