Macropixel-Constrained Collocated Positions and Neighbors Motion Search for Plenoptic 2.0 Video Coding

초록

Plenoptic 2.0 cameras enable high-resolution light field acquisition by incorporating focused optical designs that differ fundamentally from traditional plenoptic 1.0 systems. These structural differences introduce unique motion characteristics that challenge existing motion estimation (ME) algorithms. In this paper, we present comprehensive statistical analysis on real-world captured datasets to examine the primary differences in motion vector distributions among conventional, plenoptic 1.0, and plenoptic 2.0 videos. Based on the analysis, a novel fast ME algorithm is designed for efficient plenoptic 2.0 video coding. It performs a joint search over macropixel collocated positions (MCPs) and their neighboring regions to effectively accommodate the large motion deviation characteristic of plenoptic 2.0 video. To further improve computational efficiency, a macropixel-level diamond search pattern (MLDSP) is introduced to benefit from the center-biased motion-vector distribution at the macropixel resolution. Additionally, a fast MCP neighbor search is employed to restrict the search only to the top $K$ number of MCPs with the lowest distortion costs. Experimental results demonstrate that the proposed method achieves better bitrate savings and reduced computational complexity compared to existing ME techniques.

키워드