Open Collections

Mehajabin, Nusrat

University of British Columbia

Light field (LF) technology has transformed digital media by capturing scenes from multiple viewpoints, introducing a new level of visual immersion and interactivity. This thesis explores the complexities of LF technology, with a specific focus on compression techniques, quality metrics for assessing processed light fields, and refocusing strategies aligned with the principles of the human visual system. Efficient compression plays a crucial role in LF technology due to the inherent richness of data. Within this thesis, we introduce innovative pseudo-sequence-based prediction structures and coding orders for LF compression. The first structure, informed by interview-similarity, optimizes compression by leveraging reference adjacency and maximizing B-frame usage. This approach excels in bitrate efficiency, real-time decoding, and competitive encoding times, making it suitable for delay-tolerant applications such as broadcasting. The second compression technique employs a frame distance aware structure, enhancing random-access efficiency through diagonal references. This structure offers the shortest encoding time with competitive bitrate gains, making it suitable for real-time interactive applications. Existing LF refocusing methods often fall short of mimicking the natural behavior of the human visual system, resulting in refocused images that appear artificial. Therefore, we developed an innovative LF refocusing technique aimed at producing human perception consistent refocused LFs from camera arrays. By integrating view synthesis, depth estimation, and object segmentation, the proposed method addresses the challenges of depth-dependent refocusing, yielding authentic-looking, natural post-shoot refocusing for camera-array-based content. Traditional quality assessment metrics prove inadequate in capturing the multidimensional nature of LF, underscoring the necessity for novel quality metrics tailored to LF. Finally, we introduce the first LF quality metric tailored for sparsely sampled LFs. This novel full-reference LF quality assessment technique utilizes a volumetric LF representation's cross-section for holistic analysis employing a deep feature extractor. By simultaneously extracting spatial and angular features, this approach comprehensively captures LF quality. This thesis navigates the potential of LF by examining representation, compression, processing, and quality assessment. As LF continues to shape the future of digital media, the solutions presented herein facilitate its effective utilization.

Text

2024-07-10

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/88602

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/