With the fast development and application of virtual reality (VR), panoramic images have been a hot research topic and been widely applied in many areas. Usually, the panoramic images are generated by image stitching with images captured by cameras set in 360 degrees. However, the image captured by the traditional fish-eye lens has serious distortion. Many traditional image fusion algorithms are not suitable for this kind of images. In order to handle this problem, we propose a method based on multi-band image blending to realize the real-time stitching of six-way fish-eye video, and finally generates output video. The principle of the stitching system is to decompose the video into frames. After image alignment, multi-band fusion method is used to fuse the high-frequency features, which makes the degree of fusion in the image area smaller and avoids duplication in the mosaic area. Different from the convolution pyramid stitching algorithm which does not depend too much on the overlap area, the proposed algorithm retains more image details and reduces the distortion. Experimental results in terms of both subjective and objective quality show that the system can stitch panoramic video efficiently, and the output image has no stitching artifact in the overlapping area.

Date Published: 6 May 2019
PDF: 10 pages
Proc. SPIE 11069, Tenth International Conference on Graphics and Image Processing (ICGIP 2018), 110690F (6 May 2019); doi: 10.1117/12.2524247

Published in SPIE Proceedings Vol. 11069:
Tenth International Conference on Graphics and Image Processing (ICGIP 2018)
Chunming Li; Hui Yu; Zhigeng Pan; Yifei Pu, Editor(s)