Applying light field (LF) stitching on adjacent LFs can obtain an LF with a wider field of view. However, existing methods hardly maintain a consistent angular relationship for stitched LFs, and they usually have high computational complexity. This paper proposes a novel LF stitching solution to address the two issues. It includes a mesh deformation for LF alignment and a low-rank approximation for the fusion of aligned LFs. To keep the angular consistency of LFs, geometry information is taken into the proposed mesh deformation so that the aligned LFs can follow the correct angular relationship. To reduce complexity, we adopt the low-rank approximation to simplify the 4D LF into 2D matrices. Thus, a simple 2D graph-cut can efficiently fuse the aligned LFs. Experimental results show that the proposed solution is superior to others regarding pixel matching and angular consistency. Meanwhile, it has a lower time complexity than others.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Light Field Stitching Via Mesh Deformation Alignment and Low-Rank-Based Fusion

  • Yilei Chen,
  • Ping An,
  • Chao Xue,
  • Xinpeng Huang

摘要

Applying light field (LF) stitching on adjacent LFs can obtain an LF with a wider field of view. However, existing methods hardly maintain a consistent angular relationship for stitched LFs, and they usually have high computational complexity. This paper proposes a novel LF stitching solution to address the two issues. It includes a mesh deformation for LF alignment and a low-rank approximation for the fusion of aligned LFs. To keep the angular consistency of LFs, geometry information is taken into the proposed mesh deformation so that the aligned LFs can follow the correct angular relationship. To reduce complexity, we adopt the low-rank approximation to simplify the 4D LF into 2D matrices. Thus, a simple 2D graph-cut can efficiently fuse the aligned LFs. Experimental results show that the proposed solution is superior to others regarding pixel matching and angular consistency. Meanwhile, it has a lower time complexity than others.