p*****l
发帖数: 184 | 1 Video: Google's Super Resolution algorithm explained in three minutes
Published May 29, 2019 | Lars Rehm
Space constraints in the thin bodies of modern smartphones mean camera
engineers are limited in terms of the size of image sensors they can use in
their designs. Manufacturers have therefore been pushing computational
imaging methods in order to improve the quality of their devices' image
output.
Google's Super Resolution algorithm is one such method. It involves shooting
a burst of raw photos every time the shutter is pressed and takes advantage
of the user's natural hand-shake, even if it is ever so slight. The pixel-
level differences between each of the frames in the burst can be used to
merge several images of the burst into an output file with optimized detail
at each pixel location.
An illustration that shows how multiple frames are aligned to create the
final image.
Google uses the Super Resolution in the Night Sight feature and Super-Res
zoom of the Pixel 3 devices and has previously published an in-depth article
about it on its blog . Our own Rishi Sanyal has also had a close look at
the technology and the features it has been implemented in.
A visual representation of the steps used to create the final image from a
burst of Raw input images.
Now Google has published the above video that provides a great overview of
the the technology in just over three minutes.
'This approach, which includes no explicit demosaicing step, serves to both
increase image resolution and boost signal to noise ratio,' write the Google
researchers in the paper the video is based on. 'Our algorithm is robust to
challenging scene conditions: local motion, occlusion, or scene changes. It
runs at 100 milliseconds per 12-megapixel RAW input burst frame on mass-
produced mobile phones.'
Source: Google |
p*****l
发帖数: 184 | |
a***e
发帖数: 27968 | 3 典型pixel shift技术但是这个技术原则上能提高有限，而且假定很多东西不动，补出
来结果就是zoom in细节比如树叶都会糊了
in
★ 发自iPhone App: ChinaWeb 1.1.5
【在 p*****l 的大作中提到】
: Video: Google's Super Resolution algorithm explained in three minutes
: Published May 29, 2019 | Lars Rehm
: Space constraints in the thin bodies of modern smartphones mean camera
: engineers are limited in terms of the size of image sensors they can use in
: their designs. Manufacturers have therefore been pushing computational
: imaging methods in order to improve the quality of their devices' image
: output.
: Google's Super Resolution algorithm is one such method. It involves shooting
: a burst of raw photos every time the shutter is pressed and takes advantage
: of the user's natural hand-shake, even if it is ever so slight. The pixel-
|
t******8
发帖数: 2803 | 4 感觉不是新技术。
5年前试图用handbrake上的super resolution routine来放大GoPro 3+拍的240fps视频
，结果惨不忍睹。可能是因为GoPro已经对图像经过巨大的压缩，再superres一下会加
剧压缩artifact。
Google在压缩编码之前做原生的super resolution还是有戏的，做temporal noise
reduction也可以，只要处理器够猛。期待 |
s****r
发帖数: 31686 | 5 都是一些图像炼金术， 希望无中生有， 哈哈哈哈哈哈
【在 t******8 的大作中提到】
: 感觉不是新技术。
: 5年前试图用handbrake上的super resolution routine来放大GoPro 3+拍的240fps视频
: ，结果惨不忍睹。可能是因为GoPro已经对图像经过巨大的压缩，再superres一下会加
: 剧压缩artifact。
: Google在压缩编码之前做原生的super resolution还是有戏的，做temporal noise
: reduction也可以，只要处理器够猛。期待
|
