1, multi-source image - multiview
Summary of super-resolution image restoration _ _ Sina blog yield true Indian http://blog.sina.com.cn/s/blog_91991d6501018cjr.html
HR image obtained from multiple LR images:
Information, so that you can not provide new information for the rehabilitation HR image. If each piece of LR images are different sub-pixel shift between each other, so between each other will not be included with each other, in this case, each an HR image LR image will be restored to provide a number of different information. In order to obtain different sides of the same scene, scene motion must be connected to the relevant multi-scene or a video sequence by one. We passed through a multi-shot camera or a plurality of scenes acquired at a plurality of cameras at different locations, for example, in a satellite-based imaging applications orbit controlled, this movement can be achieved in a scene; for topical or moving objects uncontrollable a shock applied to the image is the same class can be realized. If the scene motion are known or in the range of sub-pixel precision it may be estimated, and if we can synthesize the HR image, the SR image restoration can be achieved.
The concept of multi-frame super-resolution. LR image grid on the left represents the same scene sub-pixel alignment, and thus can complement the SR method information acquired by fusing HR image (the right side of the grid).
Image super-resolution: technology, applications and future - Jane book https://www.jianshu.com/p/a36a333e8200
2, multi-source
Multi-source image super-resolution reconstruction of http://www.wanfangdata.com.cn/details/detail.do?_type=degree&id=Y2807120#
We studied for super-resolution reconstruction of mixed-resolution multi-view images, proposed super-resolution reconstruction algorithm based on multi-view stereoscopic depth information integration. We stereo matching algorithm using stereo matching between the reference image and the associated low-resolution images to obtain parallax images, and the reference perspective image as a reference, starting from a plurality of sets of parallax images by means of a fusion process to give the desired median filter estimate the depth of the image.
多图像超分辨率的深度学习 - 知乎 https://zhuanlan.zhihu.com/p/58286815