Super-resolution in Mobile Devices

Super-resolution is a mathematical algorithm that combines multiple low-resolution input images and creates one image of higher resolution. We provide an efficient implementation that runs in mobile devices equipped with digital cameras.

Algorithm:

 
 

1) Acquire multiple images in a burst mode

2) Registration with subpixel accuracy  3) Interpolation on high-resolution grid  4) Deconvolution of the high-res. image
  • YUV (N21) format
  • registration only on the Y channel
  • optical flow with a homography constraint
  • pyramidal approach
  • mean or
  • median more robust to misregistration
  • using precomputed sensor blur
  • Wiener filter with FFTW optimized for ARM Neon

How super-resolution goes beyond the standard camera resolution?

  Input image
 

Super-resolution

from 5 images

 

Super_resolution

from 10 images

 

Supplementary material: demo video

The Android application will be available soon. You can download now a demo application (tested only on Nexus 5), which performs SR on the central region of photos.

More examples:

Move your mouse over the captured photos to see the restored images in better quality. Notice that not only noise is removed but also small details are restored, which is the result of super-resolution. By clicking on the images you can download the input and output photos in full resolution.

Compare the output with: the original jpeg image the reference "raw" input image

 

Details: funded by Artemis No. 621439, "Almarvi" and GACR No. GA13-29225S
Duration: 2014-2017
Contact person: Filip Sroubek
Involved people: Barbara Zitova, Jan Kamenický, Ondřej Šindelář

Publications:

 

Algorithms, Design Methods, and Many-Core Execution Platform for Low-Power Massive Data-Rate Video and Image Processing

Artemis No. 621439

 

AttachmentSize
Mobile SR demonstration video (srdemo.avi)10.1 MB
Mobile SR examples (mobilesrphotos.zip)16.7 MB
SRCamera Demo Application1.29 MB